Maps API

Package nject is a general purpose dependency injection framework. It provides wrapping, pruning, and indirect variable passing. It is type safe and using it requires no type assertions. There are two main injection APIs: Run and Bind. Bind is designed to be used at program initialization and does as much work as possible then rather than during main execution. The API for nject is a list of providers (injectors) that are run in order. The final function in the list must be called. The other functions are called if their value is consumed by a later function that must be called. Here is a simple example: In this example, context.Background and log.Default are not invoked because their outputs are not used by the final function (http.ListenAndServe). The basic idea of nject is to assemble a Collection of providers and then use that collection to supply inputs for functions that may use some or all of the provided types. One big win from dependency injection with nject is the ability to reshape various different functions into a single signature. For example, having a bunch of functions with different APIs all bound as http.HandlerFunc is easy. Providers produce or consume data. The data is distinguished by its type. If you want to three different strings, then define three different types: Then you can have a function that does things with the three types: The above function would be a valid injector or final function in a provider Collection. For example: This creates a sequence and executes it. Run injects a myFirst value and the sequence of providers runs: genSecond() injects a mySecond and myStringFunc() combines the myFirst and mySecond to create a myThird. Then the function given in run saves that final value. The expected output is Providers are grouped as into linear sequences. When building an injection chain, the providers are grouped into several sets: LITERAL, STATIC, RUN. The LITERAL and STATIC sets run once per initialization. The RUN set runs once per invocation. Providers within a set are executed in the order that they were originally specified. Providers whose outputs are not consumed are omitted unless they are marked Required(). Collections are bound with Bind(&invocationFunction, &initializationFunction). The invocationFunction is expected to be used over and over, but the initializationFunction is expected to be used less frequently. The STATIC set is re-invoked each time the initialization function is run. The LITERAL set is just the literal values in the collection. The STATIC set is composed of the cacheable injectors. The RUN set if everything else. All injectors have the following type signature: None of the input or output parameters may be anonymously-typed functions. An anoymously-typed function is a function without a named type. Injectors whose output values are not used by a downstream handler are dropped from the handler chain. They are not invoked. Injectors that have no output values are a special case and they are always retained in the handler chain. In injector that is annotated as Cacheable() may promoted to the STATIC set. An injector that is annotated as MustCache() must be promoted to the STATIC set: if it cannot be promoted then the collection is deemed invalid. An injector may not be promoted to the STATIC set if it takes as input data that comes from a provider that is not in the STATIC or LITERAL sets. For example, arguments to the invocation function, if the invoke function takes an int as one of its inputs, then no injector that takes an int as an argument may be promoted to the STATIC set. Injectors in the STATIC set will be run exactly once per set of input values. If the inputs are consistent, then the output will be a singleton. This is true across injection chains. If the following provider is used in multiple chains, as long as the same integer is injected, all chains will share the same pointer. Injectors in the STATIC set are only run for initialization. For some things, like opening a database, that may still be too often. Injectors that are marked Memoized must be promoted to the static set. Memoized injectors are only run once per combination of inputs. Their outputs are remembered. If called enough times with different arguments, memory will be exhausted. Memoized injectors may not have more than 90 inputs. Memoized injectors may not have any inputs that are go maps, slices, or functions. Arrays, structs, and interfaces are okay. This requirement is recursive so a struct that that has a slice in it is not okay. Fallible injectors are special injectors that change the behavior of the injection chain if they return error. Fallible injectors in the RUN set, that return error will terminate execution of the injection chain. A non-wrapper function that returns nject.TerminalError is a fallible injector. The TerminalError does not have to be the last return value. The nject package converts TerminalError objects into error objects so only the fallible injector should use TerminalError. Anything that consumes the TerminalError should do so by consuming error instead. Fallible injectors can be in both the STATIC set and the RUN set. Their behavior is a bit different. If a non-nil value is returned as the TerminalError from a fallible injector in the RUN set, none of the downstream providers will be called. The provider chain returns from that point with the TerminalError as a return value. Since all return values must be consumed by a middleware provider or the bound invoke function, fallible injectors must come downstream from a middleware handler that takes error as a returned value if the invoke function (function that runs a bound injection chain) does not return error. If a fallible injector returns nil for the TerminalError, the other output values are made available for downstream handlers to consume. The other output values are not considered return values and are not available to be consumed by upstream middleware handlers. The error returned by a fallible injector is not available downstream. If a non-nil value is returned as the TerminalError from a fallible injector in the STATIC set, the rest of the STATIC set will be skipped. If there is an init function and it returns error, then the value returned by the fallible injector will be returned via init function. Unlike fallible injectors in the RUN set, the error output by a fallible injector in the STATIC set is available downstream (but only in the RUN set -- nothing else in the STATIC set will execute). Some examples: A wrap function interrupts the linear sequence of providers. It may or may invoke the remainder of the sequence that comes after it. The remainder of the sequence is provided to the wrap function as a function that it may call. The type signature of a wrap function is a function that receives an function as its first parameter. That function must be of an anonymous type: For example: When this wrappper function runs, it is responsible for invoking the rest of the provider chain. It does this by calling inner(). The parameters to inner are available as inputs to downstream providers. The value(s) returned by inner come from the return values of other wrapper functions and from the return value(s) of the final function. Wrap functions can call inner() zero or more times. The values returned by wrap functions must be consumed by another upstream wrap function or by the init function (if using Bind()). Wrap functions have a small amount of runtime overhead compared to other kinds of functions: one call to reflect.MakeFunc(). Wrap functions serve the same role as middleware, but are usually easier to write. Wrap functions that invoke inner() multiple times in parallel are are not well supported at this time and such invocations must have the wrap function decorated with Parallel(). Final functions are simply the last provider in the chain. They look like regular Go functions. Their input parameters come from other providers. Their return values (if any) must be consumed by an upstream wrapper function or by the init function (if using Bind()). Wrap functions that return error should take error as a returned value so that they do not mask a downstream error. Wrap functions should not return TerminalError because they internally control if the downstream chain is called. Literal values are values in the provider chain that are not functions. Provider chains can be invalid for many reasons: inputs of a type not provided earlier in the chain; annotations that cannot be honored (eg. MustCache & Memoize); return values that are not consumed; functions that take or return functions with an anymous type other than wrapper functions; A chain that does not terminate with a function; etc. Bind() and Run() will return error when presented with an invalid provider chain. Bind() and Run() will return error rather than panic. After Bind()ing an init and invoke function, calling them will not panic unless a provider panic()s A wrapper function can be used to catch panics and turn them into errors. When doing that, it is important to propagate any errors that are coming up the chain. If there is no guaranteed function that will return error, one can be added with Shun(). Bind() uses a complex and somewhat expensive O(n^2) set of rules to evaluate which providers should be included in a chain and which can be dropped. The goal is to keep the ones you want and remove the ones you don't want. Bind() tries to figure this out based on the dependencies and the annotations. MustConsume, not Desired: Only include if at least one output is transitively consumed by a Required or Desired chain element and all outputs are consumed by some other provider. Not MustConsume, not Desired: only include if at least one output is transitively consumed by a Required or Desired provider. Not MustConsume, Desired: Include if all inputs are available. MustConsume, Desired: Only include if all outputs are transitively consumed by a required or Desired chain element. When there are multiple providers of a type, Bind() tries to get it from the closest provider. Providers that have unmet dependencies will be eliminated from the chain unless they're Required. The remainder of this document consists of suggestions for how to use nject. Contributions to this section would be welcome. Also links to blogs or other discussions of using nject in practice. The best practice for using nject inside a large project is to have a few common chains that everyone imports. Most of the time, these common chains will be early in the sequence of providers. Customization of the import chains happens in many places. This is true for services, libraries, and tests. For tests, a wrapper that includes the standard chain makes it easier to write tests. See github.com/memsql/ntest for helper functions and more examples. If nject cannot bind or run a chain, it will return error. The returned error is generally very good, but it does not contain the full debugging output. The full debugging output can be obtained with the DetailedError function. If the detailed error shows that nject has a bug, note that part of the debug output includes a regression test that can be turned into an nject issue. Remove the comments to hide the original type names. The Reorder() decorator allows injection chains to be fully or partially reordered. Reorder is currently limited to a single pass and does not know which injectors are ultimately going to be included in the final chain. It is likely that if you mark your entire chain with Reorder, you'll have unexpected results. On the other hand, Reorder provides safe and easy way to solve some common problems. For example: providing optional options to an injected dependency. Because the default options are marked as Shun, they'll only be included if they have to be included. If a user of thingChain wants to override the options, they simply need to mark their override as Reorder. To make this extra friendly, a helper function to do the override can be provided and used. Recommended best practice is to have injectors shutdown the things they themselves start. They should do their own cleanup. Inside tests, an injector can use t.Cleanup() for this. For services, something like t.Cleanup can easily be built: Alternatively, any wrapper function can do it's own cleanup in a defer that it defines. Wrapper functions have a small runtime performance penalty, so if you have more than a couple of providers that need cleanup, it makes sense to include something like CleaningService. The normal direction of forced inclusion is that an upstream provider is required because a downstream provider uses a type produced by the upstream provider. There are times when the relationship needs to be reversed. For example, a type gets modified by a downstream injector. The simplest option is to combine the providers into one function. Another possibility is to mark the upstream provider with MustConsume and have it produce a type that is only consumed by the downstream provider. Lastly, the providers can be grouped with Cluster so that they'll be included or excluded as a group. Example shows what gets included and what does not for several injection chains. These examples are meant to show the subtlety of what gets included and why. This example explores injecting a database handle or transaction only when they're used.

Package skipper provides an HTTP routing library with flexible configuration as well as a runtime update of the routing rules. Skipper works as an HTTP reverse proxy that is responsible for mapping incoming requests to multiple HTTP backend services, based on routes that are selected by the request attributes. At the same time, both the requests and the responses can be augmented by a filter chain that is specifically defined for each route. Optionally, it can provide circuit breaker mechanism individually for each backend host. Skipper can load and update the route definitions from multiple data sources without being restarted. It provides a default executable command with a few built-in filters, however, its primary use case is to be extended with custom filters, predicates or data sources. For further information read 'Extending Skipper'. Skipper took the core design and inspiration from Vulcand: https://github.com/mailgun/vulcand. Skipper is 'go get' compatible. If needed, create a 'go workspace' first: Get the Skipper packages: Create a file with a route: Optionally, verify the syntax of the file: Start Skipper and make an HTTP request: The core of Skipper's request processing is implemented by a reverse proxy in the 'proxy' package. The proxy receives the incoming request, forwards it to the routing engine in order to receive the most specific matching route. When a route matches, the request is forwarded to all filters defined by it. The filters can modify the request or execute any kind of program logic. Once the request has been processed by all the filters, it is forwarded to the backend endpoint of the route. The response from the backend goes once again through all the filters in reverse order. Finally, it is mapped as the response of the original incoming request. Besides the default proxying mechanism, it is possible to define routes without a real network backend endpoint. One of these cases is called a 'shunt' backend, in which case one of the filters needs to handle the request providing its own response (e.g. the 'static' filter). Actually, filters themselves can instruct the request flow to shunt by calling the Serve(*http.Response) method of the filter context. Another case of a route without a network backend is the 'loopback'. A loopback route can be used to match a request, modified by filters, against the lookup tree with different conditions and then execute a different route. One example scenario can be to use a single route as an entry point to execute some calculation to get an A/B testing decision and then matching the updated request metadata for the actual destination route. This way the calculation can be executed for only those requests that don't contain information about a previously calculated decision. For further details, see the 'proxy' and 'filters' package documentation. Finding a request's route happens by matching the request attributes to the conditions in the route's definitions. Such definitions may have the following conditions: - method - path (optionally with wildcards) - path regular expressions - host regular expressions - headers - header regular expressions It is also possible to create custom predicates with any other matching criteria. The relation between the conditions in a route definition is 'and', meaning, that a request must fulfill each condition to match a route. For further details, see the 'routing' package documentation. Filters are applied in order of definition to the request and in reverse order to the response. They are used to modify request and response attributes, such as headers, or execute background tasks, like logging. Some filters may handle the requests without proxying them to service backends. Filters, depending on their implementation, may accept/require parameters, that are set specifically to the route. For further details, see the 'filters' package documentation. Each route has one of the following backends: HTTP endpoint, shunt or loopback. Backend endpoints can be any HTTP service. They are specified by their network address, including the protocol scheme, the domain name or the IP address, and optionally the port number: e.g. "https://www.example.org:4242". (The path and query are sent from the original request, or set by filters.) A shunt route means that Skipper handles the request alone and doesn't make requests to a backend service. In this case, it is the responsibility of one of the filters to generate the response. A loopback route executes the routing mechanism on current state of the request from the start, including the route lookup. This way it serves as a form of an internal redirect. Route definitions consist of the following: - request matching conditions (predicates) - filter chain (optional) - backend (either an HTTP endpoint or a shunt) The eskip package implements the in-memory and text representations of route definitions, including a parser. (Note to contributors: in order to stay compatible with 'go get', the generated part of the parser is stored in the repository. When changing the grammar, 'go generate' needs to be executed explicitly to update the parser.) For further details, see the 'eskip' package documentation Skipper has filter implementations of basic auth and OAuth2. It can be integrated with tokeninfo based OAuth2 providers. For details, see: https://godoc.org/github.com/zalando/skipper/filters/auth. Skipper's route definitions of Skipper are loaded from one or more data sources. It can receive incremental updates from those data sources at runtime. It provides three different data clients: - Kubernetes: Skipper can be used as part of a Kubernetes Ingress Controller implementation together with https://github.com/zalando-incubator/kube-ingress-aws-controller . In this scenario, Skipper uses the Kubernetes API's Ingress extensions as a source for routing. For a complete deployment example, see more details in: https://github.com/zalando-incubator/kubernetes-on-aws/ . - Innkeeper: the Innkeeper service implements a storage for large sets of Skipper routes, with an HTTP+JSON API, OAuth2 authentication and role management. See the 'innkeeper' package and https://github.com/zalando/innkeeper. - etcd: Skipper can load routes and receive updates from etcd clusters (https://github.com/coreos/etcd). See the 'etcd' package. - static file: package eskipfile implements a simple data client, which can load route definitions from a static file in eskip format. Currently, it loads the routes on startup. It doesn't support runtime updates. Skipper can use additional data sources, provided by extensions. Sources must implement the DataClient interface in the routing package. Skipper provides circuit breakers, configured either globally, based on backend hosts or based on individual routes. It supports two types of circuit breaker behavior: open on N consecutive failures, or open on N failures out of M requests. For details, see: https://godoc.org/github.com/zalando/skipper/circuit. Skipper can be started with the default executable command 'skipper', or as a library built into an application. The easiest way to start Skipper as a library is to execute the 'Run' function of the current, root package. Each option accepted by the 'Run' function is wired in the default executable as well, as a command line flag. E.g. EtcdUrls becomes -etcd-urls as a comma separated list. For command line help, enter: An additional utility, eskip, can be used to verify, print, update and delete routes from/to files or etcd (Innkeeper on the roadmap). See the cmd/eskip command package, and/or enter in the command line: Skipper doesn't use dynamically loaded plugins, however, it can be used as a library, and it can be extended with custom predicates, filters and/or custom data sources. To create a custom predicate, one needs to implement the PredicateSpec interface in the routing package. Instances of the PredicateSpec are used internally by the routing package to create the actual Predicate objects as referenced in eskip routes, with concrete arguments. Example, randompredicate.go: In the above example, a custom predicate is created, that can be referenced in eskip definitions with the name 'Random': To create a custom filter we need to implement the Spec interface of the filters package. 'Spec' is the specification of a filter, and it is used to create concrete filter instances, while the raw route definitions are processed. Example, hellofilter.go: The above example creates a filter specification, and in the routes where they are included, the filter instances will set the 'X-Hello' header for each and every response. The name of the filter is 'hello', and in a route definition it is referenced as: The easiest way to create a custom Skipper variant is to implement the required filters (as in the example above) by importing the Skipper package, and starting it with the 'Run' command. Example, hello.go: A file containing the routes, routes.eskip: Start the custom router: The 'Run' function in the root Skipper package starts its own listener but it doesn't provide the best composability. The proxy package, however, provides a standard http.Handler, so it is possible to use it in a more complex solution as a building block for routing. Skipper provides detailed logging of failures, and access logs in Apache log format. Skipper also collects detailed performance metrics, and exposes them on a separate listener endpoint for pulling snapshots. For details, see the 'logging' and 'metrics' packages documentation. The router's performance depends on the environment and on the used filters. Under ideal circumstances, and without filters, the biggest time factor is the route lookup. Skipper is able to scale to thousands of routes with logarithmic performance degradation. However, this comes at the cost of increased memory consumption, due to storing the whole lookup tree in a single structure. Benchmarks for the tree lookup can be run by: In case more aggressive scale is needed, it is possible to setup Skipper in a cascade model, with multiple Skipper instances for specific route segments.

Package gocql implements a fast and robust Cassandra driver for the Go programming language. Pass a list of initial node IP addresses to NewCluster to create a new cluster configuration: Port can be specified as part of the address, the above is equivalent to: It is recommended to use the value set in the Cassandra config for broadcast_address or listen_address, an IP address not a domain name. This is because events from Cassandra will use the configured IP address, which is used to index connected hosts. If the domain name specified resolves to more than 1 IP address then the driver may connect multiple times to the same host, and will not mark the node being down or up from events. Then you can customize more options (see ClusterConfig): The driver tries to automatically detect the protocol version to use if not set, but you might want to set the protocol version explicitly, as it's not defined which version will be used in certain situations (for example during upgrade of the cluster when some of the nodes support different set of protocol versions than other nodes). The driver advertises the module name and version in the STARTUP message, so servers are able to detect the version. If you use replace directive in go.mod, the driver will send information about the replacement module instead. When ready, create a session from the configuration. Don't forget to Close the session once you are done with it: CQL protocol uses a SASL-based authentication mechanism and so consists of an exchange of server challenges and client response pairs. The details of the exchanged messages depend on the authenticator used. To use authentication, set ClusterConfig.Authenticator or ClusterConfig.AuthProvider. PasswordAuthenticator is provided to use for username/password authentication: By default, PasswordAuthenticator will attempt to authenticate regardless of what implementation the server returns in its AUTHENTICATE message as its authenticator, (e.g. org.apache.cassandra.auth.PasswordAuthenticator). If you wish to restrict this you may use PasswordAuthenticator.AllowedAuthenticators: It is possible to secure traffic between the client and server with TLS. To use TLS, set the ClusterConfig.SslOpts field. SslOptions embeds *tls.Config so you can set that directly. There are also helpers to load keys/certificates from files. Warning: Due to historical reasons, the SslOptions is insecure by default, so you need to set EnableHostVerification to true if no Config is set. Most users should set SslOptions.Config to a *tls.Config. SslOptions and Config.InsecureSkipVerify interact as follows: For example: To route queries to local DC first, use DCAwareRoundRobinPolicy. For example, if the datacenter you want to primarily connect is called dc1 (as configured in the database): The driver can route queries to nodes that hold data replicas based on partition key (preferring local DC). Note that TokenAwareHostPolicy can take options such as gocql.ShuffleReplicas and gocql.NonLocalReplicasFallback. We recommend running with a token aware host policy in production for maximum performance. The driver can only use token-aware routing for queries where all partition key columns are query parameters. For example, instead of use The DCAwareRoundRobinPolicy can be replaced with RackAwareRoundRobinPolicy, which takes two parameters, datacenter and rack. Instead of dividing hosts with two tiers (local datacenter and remote datacenters) it divides hosts into three (the local rack, the rest of the local datacenter, and everything else). RackAwareRoundRobinPolicy can be combined with TokenAwareHostPolicy in the same way as DCAwareRoundRobinPolicy. Create queries with Session.Query. Query values must not be reused between different executions and must not be modified after starting execution of the query. To execute a query without reading results, use Query.Exec: Single row can be read by calling Query.Scan: Multiple rows can be read using Iter.Scanner: See Example for complete example. The driver automatically prepares DML queries (SELECT/INSERT/UPDATE/DELETE/BATCH statements) and maintains a cache of prepared statements. CQL protocol does not support preparing other query types. When using CQL protocol >= 4, it is possible to use gocql.UnsetValue as the bound value of a column. This will cause the database to ignore writing the column. The main advantage is the ability to keep the same prepared statement even when you don't want to update some fields, where before you needed to make another prepared statement. Session is safe to use from multiple goroutines, so to execute multiple concurrent queries, just execute them from several worker goroutines. Gocql provides synchronously-looking API (as recommended for Go APIs) and the queries are executed asynchronously at the protocol level. Null values are are unmarshalled as zero value of the type. If you need to distinguish for example between text column being null and empty string, you can unmarshal into *string variable instead of string. See Example_nulls for full example. The driver reuses backing memory of slices when unmarshalling. This is an optimization so that a buffer does not need to be allocated for every processed row. However, you need to be careful when storing the slices to other memory structures. When you want to save the data for later use, pass a new slice every time. A common pattern is to declare the slice variable within the scanner loop: The driver supports paging of results with automatic prefetch, see ClusterConfig.PageSize, Query.PageSize, and Query.Prefetch. It is also possible to control the paging manually with Query.PageState (this disables automatic prefetch). Manual paging is useful if you want to store the page state externally, for example in a URL to allow users browse pages in a result. You might want to sign/encrypt the paging state when exposing it externally since it contains data from primary keys. Paging state is specific to the CQL protocol version and the exact query used. It is meant as opaque state that should not be modified. If you send paging state from different query or protocol version, then the behaviour is not defined (you might get unexpected results or an error from the server). For example, do not send paging state returned by node using protocol version 3 to a node using protocol version 4. Also, when using protocol version 4, paging state between Cassandra 2.2 and 3.0 is incompatible (https://issues.apache.org/jira/browse/CASSANDRA-10880). The driver does not check whether the paging state is from the same protocol version/statement. You might want to validate yourself as this could be a problem if you store paging state externally. For example, if you store paging state in a URL, the URLs might become broken when you upgrade your cluster. Call Query.PageState(nil) to fetch just the first page of the query results. Pass the page state returned by Iter.PageState to Query.PageState of a subsequent query to get the next page. If the length of slice returned by Iter.PageState is zero, there are no more pages available (or an error occurred). Using too low values of PageSize will negatively affect performance, a value below 100 is probably too low. While Cassandra returns exactly PageSize items (except for last page) in a page currently, the protocol authors explicitly reserved the right to return smaller or larger amount of items in a page for performance reasons, so don't rely on the page having the exact count of items. See Example_paging for an example of manual paging. There are certain situations when you don't know the list of columns in advance, mainly when the query is supplied by the user. Iter.Columns, Iter.RowData, Iter.MapScan and Iter.SliceMap can be used to handle this case. See Example_dynamicColumns. The CQL protocol supports sending batches of DML statements (INSERT/UPDATE/DELETE) and so does gocql. Use Session.Batch to create a new batch and then fill-in details of individual queries. Then execute the batch with Batch.Exec. Logged batches ensure atomicity, either all or none of the operations in the batch will succeed, but they have overhead to ensure this property. Unlogged batches don't have the overhead of logged batches, but don't guarantee atomicity. Updates of counters are handled specially by Cassandra so batches of counter updates have to use CounterBatch type. A counter batch can only contain statements to update counters. For unlogged batches it is recommended to send only single-partition batches (i.e. all statements in the batch should involve only a single partition). Multi-partition batch needs to be split by the coordinator node and re-sent to correct nodes. With single-partition batches you can send the batch directly to the node for the partition without incurring the additional network hop. It is also possible to pass entire BEGIN BATCH .. APPLY BATCH statement to Query.Exec. There are differences how those are executed. BEGIN BATCH statement passed to Query.Exec is prepared as a whole in a single statement. Batch.Exec prepares individual statements in the batch. If you have variable-length batches using the same statement, using Batch.Exec is more efficient. See Example_batch for an example. Query.ScanCAS or Query.MapScanCAS can be used to execute a single-statement lightweight transaction (an INSERT/UPDATE .. IF statement) and reading its result. See example for Query.MapScanCAS. Multiple-statement lightweight transactions can be executed as a logged batch that contains at least one conditional statement. All the conditions must return true for the batch to be applied. You can use Batch.ExecCAS and Batch.MapExecCAS when executing the batch to learn about the result of the LWT. See example for Batch.MapExecCAS. Queries can be marked as idempotent. Marking the query as idempotent tells the driver that the query can be executed multiple times without affecting its result. Non-idempotent queries are not eligible for retrying nor speculative execution. Idempotent queries are retried in case of errors based on the configured RetryPolicy. Queries can be retried even before they fail by setting a SpeculativeExecutionPolicy. The policy can cause the driver to retry on a different node if the query is taking longer than a specified delay even before the driver receives an error or timeout from the server. When a query is speculatively executed, the original execution is still executing. The two parallel executions of the query race to return a result, the first received result will be returned. UDTs can be mapped (un)marshaled from/to map[string]interface{} a Go struct (or a type implementing UDTUnmarshaler, UDTMarshaler, Unmarshaler or Marshaler interfaces). For structs, cql tag can be used to specify the CQL field name to be mapped to a struct field: See Example_userDefinedTypesMap, Example_userDefinedTypesStruct, ExampleUDTMarshaler, ExampleUDTUnmarshaler. It is possible to provide observer implementations that could be used to gather metrics: CQL protocol also supports tracing of queries. When enabled, the database will write information about internal events that happened during execution of the query. You can use Query.Trace to request tracing and receive the session ID that the database used to store the trace information in system_traces.sessions and system_traces.events tables. NewTraceWriter returns an implementation of Tracer that writes the events to a writer. Gathering trace information might be essential for debugging and optimizing queries, but writing traces has overhead, so this feature should not be used on production systems with very high load unless you know what you are doing. Example_batch demonstrates how to execute a batch of statements. Example_dynamicColumns demonstrates how to handle dynamic column list. Example_marshalerUnmarshaler demonstrates how to implement a Marshaler and Unmarshaler. Example_nulls demonstrates how to distinguish between null and zero value when needed. Null values are unmarshalled as zero value of the type. If you need to distinguish for example between text column being null and empty string, you can unmarshal into *string field. Example_paging demonstrates how to manually fetch pages and use page state. See also package documentation about paging. Example_set demonstrates how to use sets. Example_userDefinedTypesMap demonstrates how to work with user-defined types as maps. See also Example_userDefinedTypesStruct and examples for UDTMarshaler and UDTUnmarshaler if you want to map to structs. Example_userDefinedTypesStruct demonstrates how to work with user-defined types as structs. See also examples for UDTMarshaler and UDTUnmarshaler if you need more control/better performance.

Package autofiber provides a FastAPI-like wrapper for the Fiber web framework. It enables automatic request parsing, validation, and OpenAPI/Swagger documentation generation. Package autofiber provides OpenAPI 3.0 specification generation for automatic API documentation. Package autofiber provides OpenAPI/Swagger documentation configuration and serving utilities. Package autofiber provides route group functionality with automatic request parsing, validation, and documentation generation. Package autofiber provides handler creation utilities for automatic request parsing, validation, and response handling. Package autofiber provides map and interface parsing utilities for converting data structures to Go structs. Package autofiber provides middleware functions for automatic request parsing, validation, and response handling. Package autofiber provides route configuration options for building APIs with automatic parsing, validation, and documentation. Package autofiber provides request parsing utilities for extracting and validating data from multiple sources. Package autofiber provides HTTP route registration methods with automatic request parsing, validation, and documentation generation. Package autofiber provides core types and configuration for the AutoFiber web framework. Package autofiber provides response validation utilities for ensuring API responses match expected schemas.

Package set implements a Set using an underlying map as an aliased type. Additionally the Set implements methods from the python set type including unions, intersections, differences and symmetric differences. Data can be fed to a set from an array or slice and additionally from an iterator. The implementation avoids allocations where possible. Unlike python, sets can only consist of comparable types. This eliminates the possibility of a 'set of sets'. The API is stable. Similarly to map, sets are not goroutine safe. This is not production code but simply a demonstration of generics and iterators. Do not import. [Python Set]: https://docs.python.org/3/library/stdtypes.html#set-types-set-frozenset Frozenset is NOT supported although it would be possible. This implementation using generics has a few problems. Map keys or sets are 'comparable' (obeys the == and != operations) whereas slices are 'cmp.Ordered' (obeys <, <=, ==, >=, > operations) and this introduces some basic incompatibilities. Attempting to sort a set using slices.Sort from the go slices package will not work as this requires the cmp.Ordered constraint. A better solution is to implement Set as a stripped down version of Map in the go source code.

Package dngn is a simple random map generation library primarily made to be used for 2D games. It features a simple API, and a couple of different means to generate maps. The easiest way to kick things off when using dngn is to simply create a Layout to represent your overall game map, which can then be manipulated or have a Generate function run on it to actually generate the content on the map.

Package types implements concrete types for the dcrwallet JSON-RPC API. When communicating via the JSON-RPC protocol, all of the commands need to be marshalled to and from the the wire in the appropriate format. This package provides data structures and primitives that are registered with dcrjson to ease this process. An overview specific to this package is provided here, however it is also instructive to read the documentation for the dcrjson package (https://godoc.org/github.com/decred/dcrd/dcrjson). The types in this package map to the required parts of the protocol as discussed in the dcrjson documention To simplify the marshalling of the requests and responses, the dcrjson.MarshalCmd and dcrjson.MarshalResponse functions may be used. They return the raw bytes ready to be sent across the wire. Unmarshalling a received Request object is a two step process: This approach is used since it provides the caller with access to the additional fields in the request that are not part of the command such as the ID. Unmarshalling a received Response object is also a two step process: As above, this approach is used since it provides the caller with access to the fields in the response such as the ID and Error. This package provides two approaches for creating a new command. This first, and preferred, method is to use one of the New<Foo>Cmd functions. This allows static compile-time checking to help ensure the parameters stay in sync with the struct definitions. The second approach is the dcrjson.NewCmd function which takes a method (command) name and variable arguments. Since this package registers all of its types with dcrjson, the function will recognize them and includes full checking to ensure the parameters are accurate according to provided method, however these checks are, obviously, run-time which means any mistakes won't be found until the code is actually executed. However, it is quite useful for user-supplied commands that are intentionally dynamic. To facilitate providing consistent help to users of the RPC server, the dcrjson package exposes the GenerateHelp and function which uses reflection on commands and notifications registered by this package, as well as the provided expected result types, to generate the final help text. In addition, the dcrjson.MethodUsageText function may be used to generate consistent one-line usage for registered commands and notifications using reflection.

Package splatplug provides a zero‑dependency, cross‑platform plugin mechanism for Go applications. Plugins are standalone Go binaries that communicate with the host (loader) over stdio using JSON‑RPC. The host launches each plugin process, performs a semver‑safe handshake (including an arbitrary map[string]string of startup config), and then invokes exported symbols as RPC calls with timeouts and crash recovery. Basic host usage: Basic plugin stub: See https://pkg.go.dev/github.com/rickcollette/splatplug for full API docs and examples. Package splatplug: core RPC types for plugin handshake and calls.

bindata converts any file into managable Go source code. Useful for embedding binary data into a go program. The file data is optionally gzip compressed before being converted to a raw byte slice. The following paragraphs cover some of the customization options which can be specified in the Config struct, which must be passed into the Translate() call. When used with the `Debug` option, the generated code does not actually include the asset data. Instead, it generates function stubs which load the data from the original file on disk. The asset API remains identical between debug and release builds, so your code will not have to change. This is useful during development when you expect the assets to change often. The host application using these assets uses the same API in both cases and will not have to care where the actual data comes from. An example is a Go webserver with some embedded, static web content like HTML, JS and CSS files. While developing it, you do not want to rebuild the whole server and restart it every time you make a change to a bit of javascript. You just want to build and launch the server once. Then just press refresh in the browser to see those changes. Embedding the assets with the `debug` flag allows you to do just that. When you are finished developing and ready for deployment, just re-invoke `go-bindata` without the `-debug` flag. It will now embed the latest version of the assets. The `NoMemCopy` option will alter the way the output file is generated. It will employ a hack that allows us to read the file data directly from the compiled program's `.rodata` section. This ensures that when we call call our generated function, we omit unnecessary memcopies. The downside of this, is that it requires dependencies on the `reflect` and `unsafe` packages. These may be restricted on platforms like AppEngine and thus prevent you from using this mode. Another disadvantage is that the byte slice we create, is strictly read-only. For most use-cases this is not a problem, but if you ever try to alter the returned byte slice, a runtime panic is thrown. Use this mode only on target platforms where memory constraints are an issue. The default behaviour is to use the old code generation method. This prevents the two previously mentioned issues, but will employ at least one extra memcopy and thus increase memory requirements. For instance, consider the following two examples: This would be the default mode, using an extra memcopy but gives a safe implementation without dependencies on `reflect` and `unsafe`: Here is the same functionality, but uses the `.rodata` hack. The byte slice returned from this example can not be written to without generating a runtime error. The NoCompress option indicates that the supplied assets are *not* GZIP compressed before being turned into Go code. The data should still be accessed through a function call, so nothing changes in the API. This feature is useful if you do not care for compression, or the supplied resource is already compressed. Doing it again would not add any value and may even increase the size of the data. The default behaviour of the program is to use compression. The keys used in the `_bindata` map are the same as the input file name passed to `go-bindata`. This includes the path. In most cases, this is not desireable, as it puts potentially sensitive information in your code base. For this purpose, the tool supplies another command line flag `-prefix`. This accepts a portion of a path name, which should be stripped off from the map keys and function names. For example, running without the `-prefix` flag, we get: Running with the `-prefix` flag, we get: With the optional Tags field, you can specify any go build tags that must be fulfilled for the output file to be included in a build. This is useful when including binary data in multiple formats, where the desired format is specified at build time with the appropriate tags. The tags are appended to a `// +build` line in the beginning of the output file and must follow the build tags syntax specified by the go tool.

Package egoscale is a mapping for with the CloudStack API (http://cloudstack.apache.org/api.html) from Go. It has been designed against the Exoscale (https://www.exoscale.com/) infrastructure but should fit other CloudStack services. To build a request, construct the adequate struct. This library expects a pointer for efficiency reasons only. The response is a struct corresponding to the data at stake. E.g. DeployVirtualMachine gives a VirtualMachine, as a pointer as well to avoid big copies. Then everything within the struct is not a pointer. Find below some examples of how egoscale may be used to interact with a CloudStack endpoint, especially Exoscale itself. If anything feels odd or unclear, please let us know: https://github.com/exoscale/egoscale/issues This example deploys a virtual machine while controlling the job status as it goes. It enables a finer control over errors, e.g. HTTP timeout, and eventually a way to kill it of (from the client side). As this library is mostly an HTTP client, you can reuse all the existing tools around it. Nota bene: when running the tests or the egoscale library via another tool, e.g. the exo cli (or the cs cli), the environment variable EXOSCALE_TRACE=prefix does the above configuration for you. As a developer using egoscale as a library, you'll find it more convenient to plug your favorite io.Writer as it's Logger. All the available APIs on the server and provided by the API Discovery plugin Security Groups provide a way to isolate traffic to VMs. Rules are added via the two Authorization commands. Security Group also implement the generic List, Get and Delete interfaces (Listable, Gettable and Deletable). See: http://docs.cloudstack.apache.org/projects/cloudstack-administration/en/stable/networking_and_traffic.html#security-groups A Zone corresponds to a Data Center. You may list them. Zone implements the Listable interface, which let you perform a list in two different ways. The first exposes the underlying CloudStack request while the second one hide them and you only manipulate the structs of your interest. An Elastic IP is a way to attach an IP address to many Virtual Machines. The API side of the story configures the external environment, like the routing. Some work is required within the machine to properly configure the interfaces. See: http://docs.cloudstack.apache.org/projects/cloudstack-administration/en/latest/networking_and_traffic.html#about-elastic-ips

Package sqlt provides a type-safe SQL builder and struct mapper using Go's text/template engine. Struct mapping is handled by the structscan(https://pkg.go.dev/github.com/go-sqlt/structscan) package. The Dest function returns a structscan.Schema[Dest], which provides a fluent API for field-based value extraction and transformation. Example: package main import ( ) var query = sqlt.All[string, Data](sqlt.Parse(` `))

Package wiremock is client for WireMock API. WireMock is a simulator for HTTP-based APIs. Some might consider it a service virtualization tool or a mock server. HTTP request: With response: Stub: The client should reset all made stubs after tests: To avoid conflicts, you can delete individual rules: You can verify if a request has been made that matches the mapping.

Package z3 provides Go bindings to the Z3 SMT Solver. The bindings are a balance between idiomatic Go and being an obvious translation from the Z3 API so you can look up Z3 APIs and find the intuitive mapping in Go. The most foreign thing to Go programmers will be error handling. Rather than return the `error` type from almost every function, the z3 package mimics Z3's API by requiring you to set an error handler callback. This error handler will be invoked whenenver an error occurs. See ErrorHandler and Context.SetErrorHandler for more information.

Package hibe implements the cryptosystem described in the paper "Hierarchical Identity Based Encyprtion with Constant Size Ciphertext" by Boneh, Boyen, and Goh. The algorithms call for us to use a group G that is bilinear, i.e, there exists a bilinear map e: G x G -> G2. However, the bn256 library uses a slightly different definition of bilinear groups: it defines it as a triple of groups (G2, G1, GT) such that there exists a bilinear map e: G2 x G1 -> GT. The paper calls this an "asymmetric bilinear group". It turns out that we are lucky. Both G2 and G1, as implemented in bn256 share the same order, and that that order (bn256.Order) happens to be a prime number p. Therefore G2 and G1 are both isomorphic to Zp. This is important for two reasons. First, the algorithm requires G to be a cyclic group. Second, this implies that G2 and G1 are isomorphic to each other. This means that as long as we are careful, we can use this library to carry out a computation that is logically equivalent to the case where G2 and G1 happen to be the same group G. For simplicity, take G = G2. In other words, choose the G used in Boneh's algorithms to be the group G2 provided by bn256. In order for this work, we need to choose a single isomorphism phi: G2 -> G1 and stick with it for all operations. Let g1 be the base of G2, and g2 be the base of G1, as provided via the APIs of bn256. We define phi as follows: phi(g1 ^ a) = g2 ^ a, for all a in Z. This is well defined because G2 is isomorphic to Zp, a cyclic group. What this means is that, if we are working with some x in G to implement the algorithm, then we must do so using g1 ^ k in G2 and g2 ^ k in G1, where g1 ^ k = x. Using this method, we can emulate the requirements of Boneh's algorithm. Furthermore, note that a marshalled G1 element is 64 bytes, whereas a marshalled G2 element is 128 bytes. Therefore, we actually switch the order of arguments to the bilinear map e so that marshalled parameters and keys are smaller (since otherwise, more elements are passed as the secone argument and therefore take up a lot of space). Note that switching the order of arguments to a bilinear map (asymmetric or otherwise) maintains bilinearity. One more thing to note is that the group, as described in the paper, is multiplicative, whereas the bn256 library uses additive notation. Keep this in mind if you ever need to read through the code.

Package iter provides basic definitions and operations related to iterators over sequences. An iterator is a function that passes successive elements of a sequence to a callback function, conventionally named yield. The function stops either when the sequence is finished or when yield returns false, indicating to stop the iteration early. This package defines Seq and Seq2 (pronounced like seek—the first syllable of sequence) as shorthands for iterators that pass 1 or 2 values per sequence element to yield: Seq2 represents a sequence of paired values, conventionally key-value or index-value pairs. Yield returns true if the iterator should continue with the next element in the sequence, false if it should stop. For instance, maps.Keys returns an iterator that produces the sequence of keys of the map m, implemented as follows: Further examples can be found in The Go Blog: Range Over Function Types. Iterator functions are most often called by a range loop, as in: Iterator functions and methods are named for the sequence being walked: The iterator method on a collection type is conventionally named All, because it iterates a sequence of all the values in the collection. For a type containing multiple possible sequences, the iterator's name can indicate which sequence is being provided: If an iterator requires additional configuration, the constructor function can take additional configuration arguments: When there are multiple possible iteration orders, the method name may indicate that order: Most iterators provide the ability to walk an entire sequence: when called, the iterator does any setup necessary to start the sequence, then calls yield on successive elements of the sequence, and then cleans up before returning. Calling the iterator again walks the sequence again. Some iterators break that convention, providing the ability to walk a sequence only once. These “single-use iterators” typically report values from a data stream that cannot be rewound to start over. Calling the iterator again after stopping early may continue the stream, but calling it again after the sequence is finished will yield no values at all. Doc comments for functions or methods that return single-use iterators should document this fact: Functions and methods that accept or return iterators should use the standard Seq or Seq2 types, to ensure compatibility with range loops and other iterator adapters. The standard iterators can be thought of as “push iterators”, which push values to the yield function. Sometimes a range loop is not the most natural way to consume values of the sequence. In this case, Pull converts a standard push iterator to a “pull iterator”, which can be called to pull one value at a time from the sequence. Pull starts an iterator and returns a pair of functions—next and stop—which return the next value from the iterator and stop it, respectively. For example: If clients do not consume the sequence to completion, they must call stop, which allows the iterator function to finish and return. As shown in the example, the conventional way to ensure this is to use defer. A few packages in the standard library provide iterator-based APIs, most notably the maps and slices packages. For example, maps.Keys returns an iterator over the keys of a map, while slices.Sorted collects the values of an iterator into a slice, sorts them, and returns the slice, so to iterate over the sorted keys of a map: Iterators provide only the values of the sequence, not any direct way to modify it. If an iterator wishes to provide a mechanism for modifying a sequence during iteration, the usual approach is to define a position type with the extra operations and then provide an iterator over positions. For example, a tree implementation might provide: And then a client could delete boring values from the tree using:

Package cbor is a modern CBOR codec (RFC 8949 & RFC 7049) with CBOR tags, Go struct tags (toarray/keyasint/omitempty), Core Deterministic Encoding, CTAP2, Canonical CBOR, float64->32->16, and duplicate map key detection. Encoding options allow "preferred serialization" by encoding integers and floats to their smallest forms (e.g. float16) when values fit. Struct tags like "keyasint", "toarray" and "omitempty" make CBOR data smaller and easier to use with structs. For example, "toarray" tag makes struct fields encode to CBOR array elements. And "keyasint" makes a field encode to an element of CBOR map with specified int key. Latest docs can be viewed at https://github.com/fxamacker/cbor#cbor-library-in-go The Quick Start guide is at https://github.com/fxamacker/cbor#quick-start Function signatures identical to encoding/json include: Standard interfaces include: Custom encoding and decoding is possible by implementing standard interfaces for user-defined Go types. Codec functions are available at package-level (using defaults options) or by creating modes from options at runtime. "Mode" in this API means definite way of encoding (EncMode) or decoding (DecMode). EncMode and DecMode interfaces are created from EncOptions or DecOptions structs. Modes use immutable options to avoid side-effects and simplify concurrency. Behavior of modes won't accidentally change at runtime after they're created. Modes are intended to be reused and are safe for concurrent use. EncMode and DecMode Interfaces Using Default Encoding Mode Using Default Decoding Mode Creating and Using Encoding Modes Predefined Encoding Options: https://github.com/fxamacker/cbor#predefined-encoding-options Encoding Options: https://github.com/fxamacker/cbor#encoding-options Decoding Options: https://github.com/fxamacker/cbor#decoding-options Struct tags like `cbor:"name,omitempty"` and `json:"name,omitempty"` work as expected. If both struct tags are specified then `cbor` is used. Struct tags like "keyasint", "toarray", and "omitempty" make it easy to use very compact formats like COSE and CWT (CBOR Web Tokens) with structs. For example, "toarray" makes struct fields encode to array elements. And "keyasint" makes struct fields encode to elements of CBOR map with int keys. https://raw.githubusercontent.com/fxamacker/images/master/cbor/v2.0.0/cbor_easy_api.png Struct tags are listed at https://github.com/fxamacker/cbor#struct-tags-1 Over 375 tests are included in this package. Cover-guided fuzzing is handled by fxamacker/cbor-fuzz.

Package goBolt implements drivers for the Neo4J Bolt Protocol Versions 1-4. There are some limitations to the types of collections the internalDriver supports. Specifically, maps should always be of type map[string]interface{} and lists should always be of type []interface{}. It doesn't seem that the Bolt protocol supports uint64 either, so the biggest number it can send right now is the int64 max. The URL format is: `bolt://(user):(password)@(host):(port)` Schema must be `bolt`. User and password is only necessary if you are authenticating. TLS is supported by using query parameters on the connection string, like so: `bolt://host:port?tls=true&tls_no_verify=false` The supported query params are: * timeout - the number of seconds to set the connection timeout to. Defaults to 60 seconds. * tls - Set to 'true' or '1' if you want to use TLS encryption * tls_no_verify - Set to 'true' or '1' if you want to accept any server certificate (for testing, not secure) * tls_ca_cert_file - path to a custom ca cert for a self-signed TLS cert * tls_cert_file - path to a cert file for this client (need to verify this is processed by Neo4j) * tls_key_file - path to a key file for this client (need to verify this is processed by Neo4j) Errors returned from the API support wrapping, so if you receive an error from the library, it might be wrapping other errors. You can get the innermost error by using the `InnerMost` method. Failure messages from Neo4J are reported, along with their metadata, as an error. In order to get the failure message metadata from a wrapped error, you can do so by calling `err.(*errors.Error).InnerMost().(messages.FailureMessage).Metadata` If there is an error with the database connection, you should get a sql/internalDriver ErrBadConn as per the best practice recommendations of the Golang SQL Driver. However, this error may be wrapped, so you might have to call `InnerMost` to get it, as specified above.

Package nject is a general purpose dependency injection framework. It provides wrapping, pruning, and indirect variable passing. It is type safe and using it requires no type assertions. There are two main injection APIs: Run and Bind. Bind is designed to be used at program initialization and does as much work as possible then rather than during main execution. The API for nject is a list of providers (injectors) that are run in order. The final function in the list must be called. The other functions are called if their value is consumed by a later function that must be called. Here is a simple example: In this example, context.Background and log.Default are not invoked because their outputs are not used by the final function (http.ListenAndServe). The basic idea of nject is to assemble a Collection of providers and then use that collection to supply inputs for functions that may use some or all of the provided types. One big win from dependency injection with nject is the ability to reshape various different functions into a single signature. For example, having a bunch of functions with different APIs all bound as http.HandlerFunc is easy. Providers produce or consume data. The data is distinguished by its type. If you want to three different strings, then define three different types: Then you can have a function that does things with the three types: The above function would be a valid injector or final function in a provider Collection. For example: This creates a sequence and executes it. Run injects a myFirst value and the sequence of providers runs: genSecond() injects a mySecond and myStringFunc() combines the myFirst and mySecond to create a myThird. Then the function given in run saves that final value. The expected output is Providers are grouped as into linear sequences. When building an injection chain, the providers are grouped into several sets: LITERAL, STATIC, RUN. The LITERAL and STATIC sets run once per initialization. The RUN set runs once per invocation. Providers within a set are executed in the order that they were originally specified. Providers whose outputs are not consumed are omitted unless they are marked Required(). Collections are bound with Bind(&invocationFunction, &initializationFunction). The invocationFunction is expected to be used over and over, but the initializationFunction is expected to be used less frequently. The STATIC set is re-invoked each time the initialization function is run. The LITERAL set is just the literal values in the collection. The STATIC set is composed of the cacheable injectors. The RUN set if everything else. All injectors have the following type signature: None of the input or output parameters may be anonymously-typed functions. An anoymously-typed function is a function without a named type. Injectors whose output values are not used by a downstream handler are dropped from the handler chain. They are not invoked. Injectors that have no output values are a special case and they are always retained in the handler chain. In injector that is annotated as Cacheable() may promoted to the STATIC set. An injector that is annotated as MustCache() must be promoted to the STATIC set: if it cannot be promoted then the collection is deemed invalid. An injector may not be promoted to the STATIC set if it takes as input data that comes from a provider that is not in the STATIC or LITERAL sets. For example, arguments to the invocation function, if the invoke function takes an int as one of its inputs, then no injector that takes an int as an argument may be promoted to the STATIC set. Injectors in the STATIC set will be run exactly once per set of input values. If the inputs are consistent, then the output will be a singleton. This is true across injection chains. If the following provider is used in multiple chains, as long as the same integer is injected, all chains will share the same pointer. Injectors in the STATIC set are only run for initialization. For some things, like opening a database, that may still be too often. Injectors that are marked Memoized must be promoted to the static set. Memoized injectors are only run once per combination of inputs. Their outputs are remembered. If called enough times with different arguments, memory will be exhausted. Memoized injectors may not have more than 90 inputs. Memoized injectors may not have any inputs that are go maps, slices, or functions. Arrays, structs, and interfaces are okay. This requirement is recursive so a struct that that has a slice in it is not okay. Fallible injectors are special injectors that change the behavior of the injection chain if they return error. Fallible injectors in the RUN set, that return error will terminate execution of the injection chain. A non-wrapper function that returns nject.TerminalError is a fallible injector. The TerminalError does not have to be the last return value. The nject package converts TerminalError objects into error objects so only the fallible injector should use TerminalError. Anything that consumes the TerminalError should do so by consuming error instead. Fallible injectors can be in both the STATIC set and the RUN set. Their behavior is a bit different. If a non-nil value is returned as the TerminalError from a fallible injector in the RUN set, none of the downstream providers will be called. The provider chain returns from that point with the TerminalError as a return value. Since all return values must be consumed by a middleware provider or the bound invoke function, fallible injectors must come downstream from a middleware handler that takes error as a returned value if the invoke function (function that runs a bound injection chain) does not return error. If a fallible injector returns nil for the TerminalError, the other output values are made available for downstream handlers to consume. The other output values are not considered return values and are not available to be consumed by upstream middleware handlers. The error returned by a fallible injector is not available downstream. If a non-nil value is returned as the TerminalError from a fallible injector in the STATIC set, the rest of the STATIC set will be skipped. If there is an init function and it returns error, then the value returned by the fallible injector will be returned via init function. Unlike fallible injectors in the RUN set, the error output by a fallible injector in the STATIC set is available downstream (but only in the RUN set -- nothing else in the STATIC set will execute). Some examples: A wrap function interrupts the linear sequence of providers. It may or may invoke the remainder of the sequence that comes after it. The remainder of the sequence is provided to the wrap function as a function that it may call. The type signature of a wrap function is a function that receives an function as its first parameter. That function must be of an anonymous type: For example: When this wrappper function runs, it is responsible for invoking the rest of the provider chain. It does this by calling inner(). The parameters to inner are available as inputs to downstream providers. The value(s) returned by inner come from the return values of other wrapper functions and from the return value(s) of the final function. Wrap functions can call inner() zero or more times. The values returned by wrap functions must be consumed by another upstream wrap function or by the init function (if using Bind()). Wrap functions have a small amount of runtime overhead compared to other kinds of functions: one call to reflect.MakeFunc(). Wrap functions serve the same role as middleware, but are usually easier to write. Wrap functions that invoke inner() multiple times in parallel are are not well supported at this time and such invocations must have the wrap function decorated with Parallel(). Final functions are simply the last provider in the chain. They look like regular Go functions. Their input parameters come from other providers. Their return values (if any) must be consumed by an upstream wrapper function or by the init function (if using Bind()). Wrap functions that return error should take error as a returned value so that they do not mask a downstream error. Wrap functions should not return TerminalError because they internally control if the downstream chain is called. Literal values are values in the provider chain that are not functions. Provider chains can be invalid for many reasons: inputs of a type not provided earlier in the chain; annotations that cannot be honored (eg. MustCache & Memoize); return values that are not consumed; functions that take or return functions with an anymous type other than wrapper functions; A chain that does not terminate with a function; etc. Bind() and Run() will return error when presented with an invalid provider chain. Bind() and Run() will return error rather than panic. After Bind()ing an init and invoke function, calling them will not panic unless a provider panic()s A wrapper function can be used to catch panics and turn them into errors. When doing that, it is important to propagate any errors that are coming up the chain. If there is no guaranteed function that will return error, one can be added with Shun(). Bind() uses a complex and somewhat expensive O(n^2) set of rules to evaluate which providers should be included in a chain and which can be dropped. The goal is to keep the ones you want and remove the ones you don't want. Bind() tries to figure this out based on the dependencies and the annotations. MustConsume, not Desired: Only include if at least one output is transitively consumed by a Required or Desired chain element and all outputs are consumed by some other provider. Not MustConsume, not Desired: only include if at least one output is transitively consumed by a Required or Desired provider. Not MustConsume, Desired: Include if all inputs are available. MustConsume, Desired: Only include if all outputs are transitively consumed by a required or Desired chain element. When there are multiple providers of a type, Bind() tries to get it from the closest provider. Providers that have unmet dependencies will be eliminated from the chain unless they're Required. The remainder of this document consists of suggestions for how to use nject. Contributions to this section would be welcome. Also links to blogs or other discussions of using nject in practice. The best practice for using nject inside a large project is to have a few common chains that everyone imports. Most of the time, these common chains will be early in the sequence of providers. Customization of the import chains happens in many places. This is true for services, libraries, and tests. For tests, a wrapper that includes the standard chain makes it easier to write tests. See github.com/memsql/ntest for helper functions and more examples. If nject cannot bind or run a chain, it will return error. The returned error is generally very good, but it does not contain the full debugging output. The full debugging output can be obtained with the DetailedError function. If the detailed error shows that nject has a bug, note that part of the debug output includes a regression test that can be turned into an nject issue. Remove the comments to hide the original type names. The Reorder() decorator allows injection chains to be fully or partially reordered. Reorder is currently limited to a single pass and does not know which injectors are ultimately going to be included in the final chain. It is likely that if you mark your entire chain with Reorder, you'll have unexpected results. On the other hand, Reorder provides safe and easy way to solve some common problems. For example: providing optional options to an injected dependency. Because the default options are marked as Shun, they'll only be included if they have to be included. If a user of thingChain wants to override the options, they simply need to mark their override as Reorder. To make this extra friendly, a helper function to do the override can be provided and used. Recommended best practice is to have injectors shutdown the things they themselves start. They should do their own cleanup. Inside tests, an injector can use t.Cleanup() for this. For services, something like t.Cleanup can easily be built: Alternatively, any wrapper function can do it's own cleanup in a defer that it defines. Wrapper functions have a small runtime performance penalty, so if you have more than a couple of providers that need cleanup, it makes sense to include something like CleaningService. The normal direction of forced inclusion is that an upstream provider is required because a downstream provider uses a type produced by the upstream provider. There are times when the relationship needs to be reversed. For example, a type gets modified by a downstream injector. The simplest option is to combine the providers into one function. Another possibility is to mark the upstream provider with MustConsume and have it produce a type that is only consumed by the downstream provider. Lastly, the providers can be grouped with Cluster so that they'll be included or excluded as a group. Example shows what gets included and what does not for several injection chains. These examples are meant to show the subtlety of what gets included and why. This example explores injecting a database handle or transaction only when they're used.

Package zdb provides a nice API to interact with SQL databases in Go. All query functions (Exec, NumRows, InsertID Get, Select, Query) use named parameters (:name)used if params contains a map or struct; positional parameters (? or $1) are used if it doesn't. You can add multiple structs or maps, but mixing named and positional parameters is not allowed. Everything between {{:name ..}} is parsed as a conditional; for example {{:foo query}} will only be added if "foo" from params is true or not a zero type. Conditionals only work with named parameters. If the query starts with "load:" then it's loaded from the filesystem or embedded files; see Load() for details. Additional DumpArgs can be added to "dump" information to stderr for testing and debugging: Running the query twice for a select is usually safe (just slower), but running insert, update, or delete twice may cause problems.

Package gocql implements a fast and robust Cassandra driver for the Go programming language. Pass a list of initial node IP addresses to NewCluster to create a new cluster configuration: Port can be specified as part of the address, the above is equivalent to: It is recommended to use the value set in the Cassandra config for broadcast_address or listen_address, an IP address not a domain name. This is because events from Cassandra will use the configured IP address, which is used to index connected hosts. If the domain name specified resolves to more than 1 IP address then the driver may connect multiple times to the same host, and will not mark the node being down or up from events. Then you can customize more options (see ClusterConfig): The driver tries to automatically detect the protocol version to use if not set, but you might want to set the protocol version explicitly, as it's not defined which version will be used in certain situations (for example during upgrade of the cluster when some of the nodes support different set of protocol versions than other nodes). The driver advertises the module name and version in the STARTUP message, so servers are able to detect the version. If you use replace directive in go.mod, the driver will send information about the replacement module instead. When ready, create a session from the configuration. Don't forget to Close the session once you are done with it: CQL protocol uses a SASL-based authentication mechanism and so consists of an exchange of server challenges and client response pairs. The details of the exchanged messages depend on the authenticator used. To use authentication, set ClusterConfig.Authenticator or ClusterConfig.AuthProvider. PasswordAuthenticator is provided to use for username/password authentication: It is possible to secure traffic between the client and server with TLS. To use TLS, set the ClusterConfig.SslOpts field. SslOptions embeds *tls.Config so you can set that directly. There are also helpers to load keys/certificates from files. Warning: Due to historical reasons, the SslOptions is insecure by default, so you need to set EnableHostVerification to true if no Config is set. Most users should set SslOptions.Config to a *tls.Config. SslOptions and Config.InsecureSkipVerify interact as follows: For example: To route queries to local DC first, use DCAwareRoundRobinPolicy. For example, if the datacenter you want to primarily connect is called dc1 (as configured in the database): The driver can route queries to nodes that hold data replicas based on partition key (preferring local DC). Note that TokenAwareHostPolicy can take options such as gocql.ShuffleReplicas and gocql.NonLocalReplicasFallback. We recommend running with a token aware host policy in production for maximum performance. The driver can only use token-aware routing for queries where all partition key columns are query parameters. For example, instead of use The DCAwareRoundRobinPolicy can be replaced with RackAwareRoundRobinPolicy, which takes two parameters, datacenter and rack. Instead of dividing hosts with two tiers (local datacenter and remote datacenters) it divides hosts into three (the local rack, the rest of the local datacenter, and everything else). RackAwareRoundRobinPolicy can be combined with TokenAwareHostPolicy in the same way as DCAwareRoundRobinPolicy. Create queries with Session.Query. Query values must not be reused between different executions and must not be modified after starting execution of the query. To execute a query without reading results, use Query.Exec: Single row can be read by calling Query.Scan: Multiple rows can be read using Iter.Scanner: See Example for complete example. The driver automatically prepares DML queries (SELECT/INSERT/UPDATE/DELETE/BATCH statements) and maintains a cache of prepared statements. CQL protocol does not support preparing other query types. When using CQL protocol >= 4, it is possible to use gocql.UnsetValue as the bound value of a column. This will cause the database to ignore writing the column. The main advantage is the ability to keep the same prepared statement even when you don't want to update some fields, where before you needed to make another prepared statement. Session is safe to use from multiple goroutines, so to execute multiple concurrent queries, just execute them from several worker goroutines. Gocql provides synchronously-looking API (as recommended for Go APIs) and the queries are executed asynchronously at the protocol level. Null values are are unmarshalled as zero value of the type. If you need to distinguish for example between text column being null and empty string, you can unmarshal into *string variable instead of string. See Example_nulls for full example. The driver reuses backing memory of slices when unmarshalling. This is an optimization so that a buffer does not need to be allocated for every processed row. However, you need to be careful when storing the slices to other memory structures. When you want to save the data for later use, pass a new slice every time. A common pattern is to declare the slice variable within the scanner loop: The driver supports paging of results with automatic prefetch, see ClusterConfig.PageSize, Session.SetPrefetch, Query.PageSize, and Query.Prefetch. It is also possible to control the paging manually with Query.PageState (this disables automatic prefetch). Manual paging is useful if you want to store the page state externally, for example in a URL to allow users browse pages in a result. You might want to sign/encrypt the paging state when exposing it externally since it contains data from primary keys. Paging state is specific to the CQL protocol version and the exact query used. It is meant as opaque state that should not be modified. If you send paging state from different query or protocol version, then the behaviour is not defined (you might get unexpected results or an error from the server). For example, do not send paging state returned by node using protocol version 3 to a node using protocol version 4. Also, when using protocol version 4, paging state between Cassandra 2.2 and 3.0 is incompatible (https://issues.apache.org/jira/browse/CASSANDRA-10880). The driver does not check whether the paging state is from the same protocol version/statement. You might want to validate yourself as this could be a problem if you store paging state externally. For example, if you store paging state in a URL, the URLs might become broken when you upgrade your cluster. Call Query.PageState(nil) to fetch just the first page of the query results. Pass the page state returned by Iter.PageState to Query.PageState of a subsequent query to get the next page. If the length of slice returned by Iter.PageState is zero, there are no more pages available (or an error occurred). Using too low values of PageSize will negatively affect performance, a value below 100 is probably too low. While Cassandra returns exactly PageSize items (except for last page) in a page currently, the protocol authors explicitly reserved the right to return smaller or larger amount of items in a page for performance reasons, so don't rely on the page having the exact count of items. See Example_paging for an example of manual paging. There are certain situations when you don't know the list of columns in advance, mainly when the query is supplied by the user. Iter.Columns, Iter.RowData, Iter.MapScan and Iter.SliceMap can be used to handle this case. See Example_dynamicColumns. The CQL protocol supports sending batches of DML statements (INSERT/UPDATE/DELETE) and so does gocql. Use Session.NewBatch to create a new batch and then fill-in details of individual queries. Then execute the batch with Session.ExecuteBatch. Logged batches ensure atomicity, either all or none of the operations in the batch will succeed, but they have overhead to ensure this property. Unlogged batches don't have the overhead of logged batches, but don't guarantee atomicity. Updates of counters are handled specially by Cassandra so batches of counter updates have to use CounterBatch type. A counter batch can only contain statements to update counters. For unlogged batches it is recommended to send only single-partition batches (i.e. all statements in the batch should involve only a single partition). Multi-partition batch needs to be split by the coordinator node and re-sent to correct nodes. With single-partition batches you can send the batch directly to the node for the partition without incurring the additional network hop. It is also possible to pass entire BEGIN BATCH .. APPLY BATCH statement to Query.Exec. There are differences how those are executed. BEGIN BATCH statement passed to Query.Exec is prepared as a whole in a single statement. Session.ExecuteBatch prepares individual statements in the batch. If you have variable-length batches using the same statement, using Session.ExecuteBatch is more efficient. See Example_batch for an example. Query.ScanCAS or Query.MapScanCAS can be used to execute a single-statement lightweight transaction (an INSERT/UPDATE .. IF statement) and reading its result. See example for Query.MapScanCAS. Multiple-statement lightweight transactions can be executed as a logged batch that contains at least one conditional statement. All the conditions must return true for the batch to be applied. You can use Session.ExecuteBatchCAS and Session.MapExecuteBatchCAS when executing the batch to learn about the result of the LWT. See example for Session.MapExecuteBatchCAS. Queries can be marked as idempotent. Marking the query as idempotent tells the driver that the query can be executed multiple times without affecting its result. Non-idempotent queries are not eligible for retrying nor speculative execution. Idempotent queries are retried in case of errors based on the configured RetryPolicy. Queries can be retried even before they fail by setting a SpeculativeExecutionPolicy. The policy can cause the driver to retry on a different node if the query is taking longer than a specified delay even before the driver receives an error or timeout from the server. When a query is speculatively executed, the original execution is still executing. The two parallel executions of the query race to return a result, the first received result will be returned. UDTs can be mapped (un)marshaled from/to map[string]interface{} a Go struct (or a type implementing UDTUnmarshaler, UDTMarshaler, Unmarshaler or Marshaler interfaces). For structs, cql tag can be used to specify the CQL field name to be mapped to a struct field: See Example_userDefinedTypesMap, Example_userDefinedTypesStruct, ExampleUDTMarshaler, ExampleUDTUnmarshaler. It is possible to provide observer implementations that could be used to gather metrics: CQL protocol also supports tracing of queries. When enabled, the database will write information about internal events that happened during execution of the query. You can use Query.Trace to request tracing and receive the session ID that the database used to store the trace information in system_traces.sessions and system_traces.events tables. NewTraceWriter returns an implementation of Tracer that writes the events to a writer. Gathering trace information might be essential for debugging and optimizing queries, but writing traces has overhead, so this feature should not be used on production systems with very high load unless you know what you are doing. Example_batch demonstrates how to execute a batch of statements. Example_dynamicColumns demonstrates how to handle dynamic column list. Example_marshalerUnmarshaler demonstrates how to implement a Marshaler and Unmarshaler. Example_nulls demonstrates how to distinguish between null and zero value when needed. Null values are unmarshalled as zero value of the type. If you need to distinguish for example between text column being null and empty string, you can unmarshal into *string field. Example_paging demonstrates how to manually fetch pages and use page state. See also package documentation about paging. Example_set demonstrates how to use sets. Example_userDefinedTypesMap demonstrates how to work with user-defined types as maps. See also Example_userDefinedTypesStruct and examples for UDTMarshaler and UDTUnmarshaler if you want to map to structs. Example_userDefinedTypesStruct demonstrates how to work with user-defined types as structs. See also examples for UDTMarshaler and UDTUnmarshaler if you need more control/better performance.

bindata converts any file into managable Go source code. Useful for embedding binary data into a go program. The file data is optionally gzip compressed before being converted to a raw byte slice. The following paragraphs cover some of the customization options which can be specified in the Config struct, which must be passed into the Translate() call. When used with the `Debug` option, the generated code does not actually include the asset data. Instead, it generates function stubs which load the data from the original file on disk. The asset API remains identical between debug and release builds, so your code will not have to change. This is useful during development when you expect the assets to change often. The host application using these assets uses the same API in both cases and will not have to care where the actual data comes from. An example is a Go webserver with some embedded, static web content like HTML, JS and CSS files. While developing it, you do not want to rebuild the whole server and restart it every time you make a change to a bit of javascript. You just want to build and launch the server once. Then just press refresh in the browser to see those changes. Embedding the assets with the `debug` flag allows you to do just that. When you are finished developing and ready for deployment, just re-invoke `go-bindata` without the `-debug` flag. It will now embed the latest version of the assets. The `NoMemCopy` option will alter the way the output file is generated. It will employ a hack that allows us to read the file data directly from the compiled program's `.rodata` section. This ensures that when we call call our generated function, we omit unnecessary memcopies. The downside of this, is that it requires dependencies on the `reflect` and `unsafe` packages. These may be restricted on platforms like AppEngine and thus prevent you from using this mode. Another disadvantage is that the byte slice we create, is strictly read-only. For most use-cases this is not a problem, but if you ever try to alter the returned byte slice, a runtime panic is thrown. Use this mode only on target platforms where memory constraints are an issue. The default behavior is to use the old code generation method. This prevents the two previously mentioned issues, but will employ at least one extra memcopy and thus increase memory requirements. For instance, consider the following two examples: This would be the default mode, using an extra memcopy but gives a safe implementation without dependencies on `reflect` and `unsafe`: Here is the same functionality, but uses the `.rodata` hack. The byte slice returned from this example can not be written to without generating a runtime error. The NoCompress option indicates that the supplied assets are *not* GZIP compressed before being turned into Go code. The data should still be accessed through a function call, so nothing changes in the API. This feature is useful if you do not care for compression, or the supplied resource is already compressed. Doing it again would not add any value and may even increase the size of the data. The default behavior of the program is to use compression. The keys used in the `_bindata` map are the same as the input file name passed to `go-bindata`. This includes the path. In most cases, this is not desireable, as it puts potentially sensitive information in your code base. For this purpose, the tool supplies another command line flag `-prefix`. This accepts a portion of a path name, which should be stripped off from the map keys and function names. For example, running without the `-prefix` flag, we get: Running with the `-prefix` flag, we get: With the optional Tags field, you can specify any go build tags that must be fulfilled for the output file to be included in a build. This is useful when including binary data in multiple formats, where the desired format is specified at build time with the appropriate tags. The tags are appended to a `// +build` line in the beginning of the output file and must follow the build tags syntax specified by the go tool.