rocket-store

Rocket-Store

Using the filesystem as a searchable database.

Rocket-Store is a high performance solution to simple data storage and retrieval. It's taking advantage of modern file system's exceptionally advanced cashing mechanisms.

It's packaged in a single file to include, with few dependencies.

Simple to use:

result = await rs.post("cars","Mercedes",{owner:"Lisa Simpson",reg:"N3RD"}); result = await rs.get("cars","*",rs._ORDER_DESC); result = await rs.delete("cars","*cede*");

Features:

• Extremely fast
• Very reliant
• Very little footprint.
• Very flexible.
• Few dependencies
• Works without configuration or setup.
• Data stored in JSON format
• Configurable
• Also available for PHP
• Has a session store module for express
• Asynchronous mutation safe

Installation

npm install rocket-store

Usages

const rs = require('rocket-store');

Rocket-Store does not require initialization:

• The storage area defaults to the OS temp dir.
• When trying to get a non existant collection, the reply is that no records were found.
• When posting to a non existant collection, it is created.

However you can set the storage area and data format to use, with the setOption function, before doing any operation on the data.

Basic terminology

Rocket-Store was made to replace a more complex database, in a setting that required a low footprint and high performance.

Rocket-Store is intended to store and retrieve records/documents, organized in collections, using a key.

Terms used:

• Collection: name of a collections of records. (Like an SQL table)
• Record: the data store. (Like an SQL row)
• Data storage area: area/directory where collections are stored. (Like SQL data base)
• Key: every record has exactly one unique key, which is the same as a file name (same restrictions) and the same wildcards used in searches.

Compare Rocket-Store, SQL and file system terms:

Rocket-Store	SQL	File system
storage area	database	data directory root
collection	table	directory
key	key	file name
record	row	file

Post

Stores a record in a collection identified by a unique key

post(string <collection>, string <key>, mixed <record> [, integer options])

Collection name to contain the records.

Key uniquely identifying the record

No path separators or wildcards etc. are allowed in collection names and keys. Illigal charakters are silently striped off.

Options

• _ADD_AUTO_INC: Add an auto incremented sequence to the beginning of the key
• _ADD_GUID: Add a Globally Unique IDentifier to the key

Returns an associative array containing the result of the operation:

• count : number of records affected (1 on succes)
• key: string containing the actual key used

If the key already exists, the record will be replaced.

If no key is given, an auto-incremented sequence is used as key.

If the function fails for any reason, an error is thrown.

Get

Find and retrieve records, in a collection.

get([string <collection> [,string <filename with wildcards> [integer <option flags]]]])

Collection to search. If no collection name is given, get will return a list of data base assets: collections and sequences etc.

Key to search for. Can be mixed with wildcards '*' and '?'. An undefined or empty key is the equivalent of '*'

Options:

• _ORDER : Results returned are ordered alphabetically ascending.
• _ORDER_DESC : Results returned are ordered alphabetically descending.
• _KEYS : Return keys only (no records)
• _COUNT : Return record count only

Return an array of

• count : number of records affected
• key : array of keys
• result : array of records

NB: wildcards are very expensive on large datasets with most filesystems. (on a regular PC with +10^7 records in the collection, it might take up to a second to retreive one record, whereas one might retrieve up to 100.000 records with an exact key match)

Delete

Delete one or more records, whos key match.

delete([string <collection> [,string <key with wildcards>]])

Collection to search. If no collection is given, THE WHOLE DATA BASE IS DELETED!

Key to search for. Can be mixed with wildcards '*' and '?'. If no key is given, THE ENTIRE COLLECTION INCLUDING SEQUENCES IS DELETED!

Return an array of

• count : number of records or collections affected

Configuring

Configuration options is an associative array, that can be parsed during require or with the options function The array can have these options:

Set data storage directory and file format to JSON

const rs = require('rocket-store'); await rs.options({ data_storage_area : "/home/rddb/webapp", data_format : rs._FORMAT_JSON, });

index name	values
data_storage_area	The directory where the database resides. The default is to use a subdirectory to the temporary directory provided by the operating system. If that doesn't work, the DOCUMENT_ROOT directory is used.
data_format	Specify which format the records are stored in. Values are: _FORMAT_NATIVE - default. and RS_FORMAT_JSON - Use JSON data format.

Examples

Storing records:

// Initialize (Not required) const rs = require('./rocket-store'); // POST a record result = await rs.post("cars", "Mercedes_Benz_GT_R", {owner: "Lisa Simpson"}); // GET a record result = await rs.get("cars", "*"); console.log(result);

The above example will output this:

{ count: 1, key: [ 'Mercedes_Benz_GT_R' ], result: [ { owner: 'Lisa Simpson' } ] }

Inserting an auto inceremented key

File names must always be unique. If you have more than one instance of a file name, you can add an auto incremented sequence to the name:

await rs.post("cars", "BMW_740li", { owner: "Greg Onslow" }, rs._ADD_AUTO_INC); await rs.post("cars", "BMW_740li", { owner: "Sam Wise" }, rs._ADD_AUTO_INC); await rs.post("cars", "BMW_740li", { owner: "Bill Bo" }, rs._ADD_AUTO_INC); result = await rs.get("cars", "*"); console.log(result);

The above will output this:

{ count: 4, key: [ '1-BMW_740li', '2-BMW_740li', '3-BMW_740li' ], result: [ { owner: 'Greg Onslow' }, { owner: 'Sam Wise' }, { owner: 'Bill Bo' } ] }

Inserting with Globally Unique IDentifier key

Another option is to add a GUID to the key. The GUID is a combination of a timestamp and a random sequence, formatet in accordance to RFC 4122 (Valid but slightly less random)

If ID's are generated more than 1 millisecond apart, they are 100% unique. If two ID's are generated at shorter intervals, the likelyhod of collission is up to 1 of 10^15.

await rs.post("cars", "BMW_740li", { owner: "Greg Onslow" }, rs._ADD_GUID); await rs.post("cars", "BMW_740li", { owner: "Sam Wise" }, rs._ADD_GUID); await rs.post("cars", "BMW_740li", { owner: "Bill Bo" }, rs._ADD_GUID); result = await rs.get("cars", "*"); console.log(result);

The above will output this:

{ count: 4, key: [ '16b4ffd8-87a0-4000-839f-ea5dd495b000-BMW_740li', '16b4ffd8-87b0-4000-8032-45d788fac000-BMW_740li', '16b4ffd8-87b0-4000-839f-95bd498f5000-BMW_740li' ], result: [ { owner: 'Greg Onslow' }, { owner: 'Sam Wise' }, { owner: 'Bill Bo' } ] }

Mass insterts

const dataset = { Gregs_BMW_740li : { owner: "Greg Onslow" }, Lisas_Mercedes_Benz_GT_R : { owner: "Lisa Simpson" }, Bills_BMW_740li : { owner: "Bill Bo" }, }; var promises = []; var ii = 0; for(let i in dataset){ ii++; promises[promises.length] = rs.post("cars", i, dataset[i]); if(ii >= 20){ ii = 0; await Promise.all(promises); } } if(promises.length > 0) await Promise.all(promises); result = await rs.get("cars", "*"); console.log(result);

The above example might output this:

{ count: 3, key:[ 'Lisas_Mercedes_Benz_GT_R', 'Gregs_BMW_740li', 'Bills_BMW_740li', ], result: [ { owner: 'Lisa Simpson' }, { owner: 'Greg Onslow' }, { owner: 'Bill Bo' }, ] }

Get records with matching keys

result = await rs.get("cars", "*BMW*");

Get list ordered by alphabetically descending keys

result = await rs.get("cars", "*BMW*",rs._ORDER_DESC);

Get list of collections and sequences

rs.get();

Delete matching records from a collection

rs.delete("cars", "*BMW*");

Delete a whole collection

rs.delete("cars");

Delete the entire database

rs.delete();

---

File system issue

This was made with node ver 11. A compromise was struck, to compensate for the immaturity of the node file system library; There is no proper glob functionality, to filter a directory search on a low level. Instead, an array of all entries is read.

This consumes a lot of memory, with a large database. There is no avoiding that, short of improving opon the node file system library. This is beyond my intentions, at this time. I hope it will be remedied by the node core team.

Since the memory will be used anyway, it is applied to improve speed on key searching, by keeping the read keys in memory between searched, as a key_cash.

A draw back of this, is that collection names are restricted to valid variable names, as well as directory names.

Another issue is that file locking is yet to be implementet in node. Therefore a time consuming locking mecahnism is implemented as symlinks.

Both solutions will hopefully be changed, as node matures.

---

Benchmarks

Benchmarks are performed with 1 million records in in a single collection.

System	Mass insert	exact key search	wildcard search	no hit	delete
Debian, i7 3rd gen, SSD	69000/sec.	87000/sec.	14,6/sec.	123000/sec.	525/sec.
Raspbarry Pi Zero	561/sec.	96/sec.	0.27/sec.	147/sec.	10.3/sec.

---

Contributions

• I appreciate all kinds of contribution.
• Don't hesitate to submit an issue report on github. But please provide a reproducible example.
• Code should look good and compact, and be covered by a test case or example.
• Please don't change the formatting style laid out, without a good reason. I know its not the most common standard, but its rather efficient one.

---

Updates

0.10.8

• removed unneeded module sanitise-filename 0.10.7
• Bug fix: Wildcard search on windows OS failed to find valid keys.

0.10.6

• Bug fix: Corupted og invalid files now returns an empty record, instead of throwing an error.

0.10.5 repository version correction

0.10.4

• Bug fix: Asynchronous integrity of records failed. Circumvent bug in fs.extra

0.10.3

• Bug fix: Options data_storage_area ignored.

0.10.2

• Data storage directory is now set immediately. An error is thrown later, if creation fails.

0.10.1

• Refactoring of get methods
• Added get flags _COUNT and _KEYS

0.9.4:

• Added Globally Unique IDentifier option to key genration. post flag: _ADD_GUID

0.9.3:

• Cash update dublicate bug fix.

0.9.2:

• Minor fixes and rewrites