overleaf/libraries/object-persistor
Andrew Rumble 71187a51ba Merge pull request #18289 from overleaf/ac-ar-eslint-return-await
Add ESLint rule @typescript-eslint/return-await to backend services

GitOrigin-RevId: 75e3e32597827fcc852e69d479515fc72e8f45e4
2024-05-27 10:22:49 +00:00
..
src Merge pull request #18289 from overleaf/ac-ar-eslint-return-await 2024-05-27 10:22:49 +00:00
test Merge pull request #18289 from overleaf/ac-ar-eslint-return-await 2024-05-27 10:22:49 +00:00
.dockerignore Merge pull request #5724 from overleaf/jpa-lean-docker-build 2021-11-09 09:04:02 +00:00
.gitignore Merge pull request #14138 from overleaf/bg-em-jpa-mj-node-18 2023-08-03 08:03:56 +00:00
.mocharc.json Create new module from overleaf/filestore persistors (#1) 2020-07-02 14:19:45 +01:00
.nvmrc Merge pull request #17870 from overleaf/bg-node-upgrade-18-20-2 2024-04-24 08:05:16 +00:00
.prettierignore Create new module from overleaf/filestore persistors (#1) 2020-07-02 14:19:45 +01:00
buildscript.txt Merge pull request #17870 from overleaf/bg-node-upgrade-18-20-2 2024-04-24 08:05:16 +00:00
index.js Create new module from overleaf/filestore persistors (#1) 2020-07-02 14:19:45 +01:00
LICENSE Create new module from overleaf/filestore persistors (#1) 2020-07-02 14:19:45 +01:00
package.json Merge pull request #18167 from overleaf/em-simpler-typescript-backend 2024-05-07 08:04:00 +00:00
README.md Merge pull request #13274 from overleaf/csh-issue-13004-cleanup-custom-code 2023-06-02 08:05:10 +00:00
tsconfig.json Merge pull request #18167 from overleaf/em-simpler-typescript-backend 2024-05-07 08:04:00 +00:00

@overleaf/object-persistor

Stores arbitrary objects in multiple backends, with support for falling back to a secondary backend if the object can't be found in the primary.

Contains a workaround within the GCS backend to allow lifecycle rules to keep objects for a set period of time from deletion, which can't currently be accomplished with GCS's own rules. (See configuration-specific notes later)

Backends available

  • S3
  • GCS
  • Filesystem (FS)

Getting started

// import the module
const ObjectPersistor = require('object-persistor')

const config = {
  // see 'Configuration' section below
}
// create a new persistor
const Persistor = ObjectPersistor(config)

Errors

Errors returned by persistor methods are all derived from OError (@overleaf/o-error.) To perform instanceof checks, you can use the Errors object from the persistor module:

const ObjectPersistor = require('object-persistor')
const { Errors } = ObjectPersistor

Methods

sendStream

async function sendStream(bucketName, key, readStream, opts = {})

Uploads a stream to the backend.

  • bucketName: The name of the bucket to upload to
  • key: The key for the uploaded object
  • readStream: The data stream to upload
  • opts (optional):
    • sourceMd5: The md5 hash of the source data, if known. The uploaded data will be compared against this and the operation will fail if it does not match. If omitted, the md5 is calculated as the data is uploaded instead, and verified against the backend.
    • contentType: The content type to write in the object metadata
    • contentEncoding: The content encoding to write in the object metadata
Notes

When using a secondary persistor, this method uploads only to the primary.

If an object already exists at the specified key, it will be overwritten.

getObjectStream

async function getObjectStream(bucketName, key, opts = {})

Retrieves a stream from the backend, for reading

  • bucketName: The name of the bucket to download from
  • key: The key for the object
  • opts (optional):
    • start, end: Downloads a byte range from the object. Specify both start and end. end is inclusive.
Returns

A stream.Readable to read the data.

Notes

When using a secondary persistor, this method will fall back to retrieving the object from the secondary if it does not exist on the primary.

getRedirectUrl

async function getRedirectUrl(bucketName, key)

Gets a signed link directly to the backend, if possible. This can be used to download the data directly, instead of proxying it.

  • bucketName: The name of the bucket to download from
  • key: The key for the object
Returns

A string containing the signed link, or null if a link cannot be generated.

Notes

In the case of null, you should fall back to getObjectStream as sometimes signed links cannot be generated.

Do not use this method if you are using a secondary persistor, as this mechanism does not check to see if the object actually exists - so cannot provide a fallback.

getObjectSize

async function getObjectSize(bucketName, key)

Returns the size of the stored data

  • bucketName: The name of the bucket to download from
  • key: The key for the object
Returns

An integer containing the size, in bytes.

Notes

When using a secondary persistor this method returns the size from the secondary persistor, if not found on the primary.

getObjectMd5Hash

async function getObjectMd5Hash(bucketName, key)

Returns the MD5 hash of the stored data

  • bucketName: The name of the bucket to download from
  • key: The key for the object
Returns

A string containing the hex representation of the MD5 hash

Notes

When using a secondary persistor this method returns the hash from the secondary persistor, if not found on the primary.

deleteFile

async function deleteFile(bucketName, key)

Deletes an object

  • bucketName: The name of the bucket to delete from
  • key: The key for the object
Notes

When using a secondary persistor, this deletes the object from both persistors.

deleteDirectory

async function deleteDirectory(bucketName, key)

Deletes a directory (all object whose keys start with the supplied key)

  • bucketName: The name of the bucket to delete from
  • key: The key prefix for the objects
Notes

When using a secondary persistor, this deletes the objects from both persistors.

directorySize

async function directorySize(bucketName, key)

Returns the size of a directory (all objects whose keys start with the supplied key)

  • bucketName: The name of the bucket to examine
  • key: The key prefix for the objects
Returns

An integer containing the size, in bytes

Notes

When using a secondary persistor, this returns the value from the secondary persistor if no objects are found on the primary.

checkIfObjectExists

async function checkIfObjectExists(bucketName, key)

Returns whether an object exists

  • bucketName: The name of the bucket to examine
  • key: The key for the object
Returns

A boolean representing whether the object exists

Notes

When using a secondary persistor, returns true if the object exists on either the primary or secondary.

copyObject

async function copyObject(bucketName, sourceKey, destKey)

Copies a object to another key, within a bucket.

  • bucketName: The name of the bucket in which to copy the object
  • sourceKey: The key for the object to be copied
  • destKey: The key to which the object should be copied
Notes

Can only copy objects within a single bucket. To copy objects in any other way, pass the stream returned from getObjectStream to sendStream

If an object already exists at the specified key, it will be overwritten.

sendFile

async function sendFile(bucketName, key, fsPath)

Uploads a file from the local disk.

  • bucketName: The name of the bucket to upload to
  • key: The key for the uploaded object
  • fsPath: The path on disk to the file for uploading
Notes

When using a secondary persistor, this method uploads only to the primary.

If an object already exists at the specified key, it will be overwritten.

This method is designed for applications which may write temporary data out to the disk before uploading.

Configuration

An object with the relevant configuration should be passed to the main function returned from the module. The object contains both common and backend-specific parameters.

Common parameters

  • backend (required): String specifying the primary persistor to use as the storage backend. Must be one of s3, gcs or fs.
  • signedUrlExpiryInMs: Time before expiry (in milliseconds) of signed URLs

FS-specific parameters

  • useSubdirectories: If true, files will be stored in subdirectories on the filesystem. By default, the directory structure is flattened and slashes in the object keys are replaced with underscores.

Notes

For the FS persistor, the bucketName should be the full path to the folder on disk where the files are stored.

S3-specific parameters

  • s3.key (required): The AWS access key ID

  • s3.secret (required): The AWS secret access key

  • s3.partSize: The part size for S3 uploads. Defaults to 100 megabytes.

  • s3.httpOptions: HTTP options passed directly to the S3 constructor.

  • s3.maxRetries: The number of times the S3 client will retry in case of an error

  • s3.endpoint: For testing - overrides the S3 endpoint to use a different service (e.g. a fake S3 server)

  • s3.pathStyle: For testing - use old path-style URLs, for services that do not support subdomain-based access

  • s3BucketCreds: A JSON-encoded string specifying different S3 credentials for accessing different buckets, in the following format. These credentials override the default ones configured in the main s3 settings:

{
  "bucketName": {
    "auth_key": "your aws access key ID",
    "auth_secret": "your aws secret access key"
  }
}

Notes

In order for server-side MD5 generation to work, uploads must be below the partSize. Otherwise a multipart upload will be used, and the S3 eTag which is used to retrieve the MD5 will not be the MD5 hash of the uploaded object. In these cases, we download the data and calculate the MD5 manually.

For verification during upload, we use S3's checksum mechanism to verify the integrity of the uploaded data, but when explicitly retrieving the md5 hash this will download the entire object if its size is above the part size.

GCS-specific parameters

GCS authentication is configured automatically via the local service account, or the GOOGLE_APPLICATION_CREDENTIALS environment variable.

  • gcs.unlockBeforeDelete: unlock an event-based hold before deleting. default false (see notes)
  • gcs.deletedBucketSuffix: if present, copy the object to a bucket with this suffix before deletion (see notes)
  • gcs.deleteConcurrency: when recursively deleting a directory, the maximum number of delete requests that will be used at once (default 50)
  • gcs.unsignedUrls: For testing - do not sign GCS download URLs
  • gcs.endpoint.apiEndpoint: For testing - specify a different GCS endpoint to use
  • gcs.endpoint.projectId: For testing - the GCS project ID to supply to the overridden backend

Notes

In order to support deletion after a period, the GCS persistor allows usage of a two-bucket system. The main bucket contains the live objects, and on delete the objects are first copied to a 'deleted' bucket, and then deleted from the main one. The 'deleted' bucket is then expected to have a lifecycle policy applied to delete objects after a set period.

In order to prevent accidental deletion from outside this mechanism, an event-based-hold can be applied by default on the main bucket. This will be unlocked after the object has been copied to the 'deleted' bucket so that the object can then be deleted from the main bucket.

Contributing

Contributions should pass lint, formatting and unit test checks. To run these, use

npm run test

There are no acceptance tests in this module, but https://github.com/overleaf/filestore/ contains a comprehensive set of acceptance tests that use this module. These should also pass, with the changes.