Skip to content

Latest commit

 

History

History
297 lines (199 loc) · 9.02 KB

CHANGELOG.md

File metadata and controls

297 lines (199 loc) · 9.02 KB
Raw

3.3.1 | 2020-01-14

  • apply a ceiling to the linear connection backoff to prevent endless sleep

3.3.0 | 2020-01-14

  • backs off new connection attempts when they fail consecutively preventing an impossible connection from entering a tight loop
  • Worker is now an event emitter, fail is emitted on job failure
  • removes a module with trouble installing (memory profiler)
  • worker now BEATs upon startup
  • processor pool starts up synchronously instead of concurrently

3.2.0 | 2019-11-07

  • Allows configuring of the client connection pool size via poolSize constructor option for Worker. (#37)

3.0.3 | 2019-05-02

  • Listen for and log connection errors instead of allowing them to be uncaught and crashing the node.js process. (#31)
  • Increase connection timeout to 10s (#30)

3.0.2 | 2019-05-02

  • Allow failing of jobs with errors that have no stack (#29)

3.0.1 | 2019-03-29

  • Add missing passing of password refernce to client

3.0.0 | 2019-02-11

  • Refactors connection handling
  • Adds a connection pool to every Client
  • Client.connect() not necessary before using connection Client.connect() now connects and disconnects to test the connection details. A connection is created in the pool when it is needed.
  • Added @private Connection class
  • Added generated jsdocs to docs/ for public API
  • Client constructor accepts poolSize parameter for size of connection pool
  • Client and Worker exported by package: const { Client, Worker } = require('faktory-worker');
  • Compatibility with faktory v0.9.5

BREAKING

  • Job method chaining: new Job().args().queue() chaining deprecated. See API Docs for update for more information. TL;DR use accessors (job.queue = '') isntead of method chains (queue().retry(1)).
  • Client.send(command, ASSERTION): Undocumented, but Client.send() accepted a second argument--an expected response. Please use Client.sendWithAssert for this behavior.

2.2.3 | 2018-11-20

  • Prevents an uncaught error thrown while successfuly reconnecting (#24)

2.2.2 | 2018-11-16

  • enables TCP KeepAlive on client sockets to prevent timeouts and disconnects

2.2.1 | 2018-11-04

  • fixes incorrect queue shuffling with no priorities—allows queues to be processed in strict ordering

2.2.0 | 2018-10-13

  • adds job builder API to client

2.1.1 | 2018-10-13

  • update all dependencies

2.1.0 | 2018-10-13

  • test against faktory v0.8.0

2.0.1 | 2018-08-15

  • fixes an issue during shutdown in which the hard shutdown timeout was never cleared
  • adds a start_delay to Worker#tick to stagger the FETCH thundering herd
  • adds a test for the quiet signal
  • stabilizes the shutdown code with process.exit

2.0.0 | 2018-08-13

  • refactors the concurrency model

BREAKING:

  • deleted Manager faktory/manager
  • deleted Processor faktory/processor

You should have no issues if you've been using the faktory export and not Manager and Processor classes directly.

1.0.1 | 2018-08-12

1.0.0 | 2018-07-21

Implemented:

  • Handle signals from server heartbeat response
  • Middleware
  • CLI
  • Heartbeat
  • Tests
  • Authentication
  • Fail jobs
  • Add'l client commands API
  • Labels

BREAKING:

  • Middleware and thunks: fixed issue where middleware did not provide the job function with the context. The job function thunk now takes the context as an arg and the job property can be accessed within. ex: faktory.register('MyJob', (...args) => ({ job }) => {})

0.10.0 | 2018-06-04

  • moves faktory-client into this repo
  • require('faktory-worker/client') export now available
  • manager now respects faktory-server heartbeat signals (quiet, terminate) so buttons clicked in the UI will now quiet and terminate the faktory work process

0.9.2 | 2018-06-04

  • fix incorrectly pathed faktory-work bin in package.json

0.9.1 | 2018-03-30

  • Fixed issues where job payload attributes assumed to be defaulted by the faktory server were not being sent by the client. The server does not default these so now faktory-client defaults these values.

0.9.0 | 2018-03-28

  • adds faktory.use to add middleware to the job execution stack (koa.js style)

0.8.1 | 2018-03-26

  • bugfix: process.exit after graceful shutdown of manager

0.8.0 | 2018-03-24

  • Upgrade faktory-client to v0.6.0
  • FAKTORY_URL is parsed by URL lib—must contain protocol tcp:// where it previously did not

0.7.0 | 2018-02-05

  • Upgrade faktory-client to v0.5.0
  • Run tests against faktory v0.7.0
  • Allow options.heartbeatInterval override (default 15s)
  • Gracefully reconnect when connections are interrupted / closed

0.6.3 | 2017-11-19

  • Allow jobs to reject and FAIL with a string or undefined (without an error object). #2

0.6.2 | 2017-11-12

  • Upgrade faktory-client to v0.4.2

0.6.1 | 2017-11-12

  • Check job timeout completion more frequently

0.6.0 | 2017-11-12

  • Upgrade faktory-client for faktory protocol version 2
  • BUGFIX: don't try to execute jobs that aren't dispatched
  • Interpret heartbeat response for quiet|terminate
  • Fix race condition in which graceful shutdown drains the pool too early and prevents the processor from ACKing of FAILing a job
  • Heartbeat now occurs only for the manager, not for every processor and connection in the pool.
  • Tests now run in parallel; faktory is not spawned by node. The faktory server must be started before running the tests. Use bin/server for convenience.

0.5.0 | 2017-11-11

  • Shuffle the queues array to prevent queue starvation

0.4.1 | 2017-11-11

  • Throwing / propagating errors during .execute() was causing some stabilities issues when shutting down the process. I believe this was interrupting with the normal work .loop() and preventing graceful shutdown. Ideally, the errors thrown in jobs are propagated to the application code in an emitter or registered callback (e.g. errorCallbacks << () => { ... }) so applications can send those errors to a reporting service. Until then, the error is logged to console.error instead of thrown. The same applies for dispatching.

0.4.0 | 2017-11-11

  • Based on the discussion in gitter, it became evident that a best practice for throughput was to create a pool of connections equal to the desired concurrency. This didn't produce fruitful results at first, but the introduction of a Processor pool within the Manager with a connection pool of TCP sockets to the faktory server produced some substantial (2x+) improvements to throughput and some pieces of the code became easier to reason about.

Prior to this work: Concurrency: 20 Duration: 4.639131743s Jobs/s: 6467 With Processor Pool: Concurrency: 20 Duration: 3.31144746s Jobs/s: 9059

  • benchmarking was done by running faktory directly, not through docker
  • Introduce Processor pool (w/ connection pool) within Manager: the number of TCP connections to the faktory server will now be equal to the concurrency setting (default 20).
  • benchmark scripts added

0.3.0 | 2017-11-01

  • Add heartbeat
  • Allow concurrency option in constructor, CLI

0.2.6 | 2017-10-28

  • .stop() waits for the in-progress job to complete before timeout
  • Upgrades faktory-client dependency to 0.2.2

0.2.5 | 2017-10-28

  • Upgrades faktory-client to 0.2.1

0.2.4 | 2017-10-28

  • Shutdown gracefully via .stop()

0.2.0 | 2017-10-28

  • Extact client code to faktory-client package
  • Add TravisCI
  • Use async/await
  • Tests for manager.js
  • add ava for tests, nyc for coverage

faktory-client (moved to this repo)

0.6.0 | 2018-03-24

  • connection URL is now parsed by the node URL module. The previous URL parsing did not allow protocols hints (tls) or passwords in the url. Because the URL lib is now used, a connection string without a protocol cannot be used. localhost:7419 and the like will not work—please use tcp://localhost:7419

0.5.0 | 2018-02-04

  • Refactor connection process: reconnect on close if not initial connection attempt
  • Improve debug logging output
  • Use assert module for response expectations
  • Parse connection URLs better (allow tcp://)

0.4.3 | 2017-11-12

  • Bugfix: in the rare case a response is dropped from the server, the code was using a variable that was not defined to log the dropped message.

0.4.2 | 2017-11-12

  • Bugfix: server now sends NULL for fetch requests when queues are empty. The code attempted to use string methods on this, expecting that it was a buffer/string.

0.4.1 | 2017-11-12

  • Test updates

0.4.0 | 2017-11-12

Breaking

  • Must provide wid in construction if the client is going to heartbeat, otherwise it will error.
  • .beat() now returns a string with either 'OK' or the state value 'quiet'|'terminate'