NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Packagist, #884 most downloaded on Packagist
RoadRunner: High-performance PHP application server and process manager written in Go and powered with plugins
Latest stable
v2025.1.15
Last release 1 months ago
21 Oct 2026
Release timing varies
gaps range from 2 weeks to 4 months
Nearly every release is documented
notes for 60 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
9 years old
272 releases · first in 2018
Drag across the ruler to compare, or tap a release to open it.
🐛 Fix: errors when workers reallocating if the unix/tcp socket transport is used: BUG, (reporter @tarampampam)
🐛 Fix: worker sometimes panics when unix/tcp sockets transport used: BUG, (reporter @tarampampam)
unix/tcp sockets transport used: BUG, (reporter @tarampampam)🐛 Fix: case-sensitive attributes for SQS BUG, (reporter @paulermo)
Your coding agent can read these notes before it upgrades. Set up the MCP server →
🐛 Fix: temporal: ActivityNotRegisteredError on the local activity workers BUG (reporter @Zylius)
ActivityNotRegisteredError on the local activity workers BUG (reporter @Zylius)✏️ Better workers statistic. Add worker state, per-worker memory usage, and worker's PID stats for the JOBS, GRPC, HTTP plugins FR
worker state, per-worker memory usage, and worker's PID stats for the JOBS, GRPC, HTTP plugins FRsdk to v2.8.0-rc.3api to v2.8.0-rc.3✏️ Return an error if the user uses the .env file, but it doesn't exist. BUG, (reporter @OO00O0O)
.env file, but it doesn't exist. BUG, (reporter @OO00O0O)watch_tick BUG (reported by: @OO00O0O)🐛 Fix: jobs pipeline shadowing if registering the pipeline with the existing name BUG
## 👀 New: - Move to the new organization
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
But if you use configuration version 2.7 you must update the jobs pipelines config. At this point we can guarantee, that no breaking changes will be i…
workers pool, worker, worker_watcher now has their own log levels. stderr/stdout logged as before at the info log level. All other messages moved to the debug log level except a few events from the worker_watcher when RR can't allocate the new worker which is moved to the warn..rr.yaml now support versions. You may safely use your old configurations w/o specifying versions. Configuration w/o version will be treated as 2.6. It is safe to use configuration w/o version or with version 2.6 with RR 2.7 because RR is able to automatically transform the old configuration.
But if you use configuration version 2.7 you must update the jobs pipelines config.
At this point we can guarantee, that no breaking changes will be introduced in the configuration w/o auto-convert from the older configuration version
For example, if we introduce a configuration update let's say in version 2.10, we will support automatic conversion from at least 2 previous versions w/o involving the user into the process. In the example case, versions 2.9 and 2.8 will be automatically converted. From our release cycle, you will have at least 3 months to update the configuration from version 2.8 and 2 months from 2.9.Version located at the top of the .rr.yaml:Compatibility matrix located here: link
version: "2.7"
# ..... PLUGINS ......
Before:
pipelines:
test-local:
driver: memory
priority: 10
prefetch: 10000
test-local-1:
driver: boltdb
priority: 10
file: "rr.db"
prefetch: 10000
test-local-2:
driver: amqp
prefetch: 10
priority: 1
queue: test-1-queue
exchange: default
exchange_type: direct
routing_key: test
exclusive: false
multiple_ack: false
requeue_on_fail: false
test-local-3:
driver: beanstalk
priority: 11
tube_priority: 1
tube: default-1
reserve_timeout: 10s
test-local-4:
driver: sqs
priority: 10
prefetch: 10
visibility_timeout: 0
wait_time_seconds: 0
queue: default
attributes:
DelaySeconds: 0
MaximumMessageSize: 262144
MessageRetentionPeriod: 345600
ReceiveMessageWaitTimeSeconds: 0
VisibilityTimeout: 30
tags:
test: "tag"
test-local-5:
driver: nats
priority: 2
prefetch: 100
subject: default
stream: foo
deliver_new: true
rate_limit: 100
delete_stream_on_stop: false
delete_after_ack: false
After:
Now, pipelines have only driver key with the configuration under the config key. We did that to uniform configuration across all drivers (like in the KV).
pipelines:
test-local:
driver: memory
config: # <------------------ NEW
priority: 10
prefetch: 10000
test-local-1:
driver: boltdb
config: # <------------------ NEW
priority: 10
file: "test-local-1-bolt.db"
prefetch: 10000
test-local-2:
driver: amqp
config: # <------------------ NEW
priority: 11
prefetch: 100
queue: test-12-queue
exchange: default
exchange_type: direct
routing_key: test
exclusive: false
multiple_ack: false
requeue_on_fail: false
test-local-3:
driver: beanstalk
config: # <------------------ NEW
priority: 11
tube_priority: 1
tube: default-2
reserve_timeout: 10s
test-local-4:
driver: sqs
config: # <------------------ NEW
priority: 10
prefetch: 10
visibility_timeout: 0
wait_time_seconds: 0
queue: default
attributes:
DelaySeconds: 0
MaximumMessageSize: 262144
MessageRetentionPeriod: 345600
ReceiveMessageWaitTimeSeconds: 0
VisibilityTimeout: 30
tags:
test: "tag"
test-local-5:
driver: nats
config: # <------------------ NEW
priority: 2
prefetch: 100
subject: default
stream: foo
deliver_new: true
rate_limit: 100
delete_stream_on_stop: false
delete_after_ack: false
Cache-Control and caching in general. In the first alpha you may test the max-age, Age, and Authorization support via the in-memory driver. (reporter: @alexander-schranz)Configuration:
http:
# .....
middleware: ["cache"]
cache:
driver: memory
cache_methods: ["GET", "HEAD", "POST"] # only GET in alpha
config: {} # empty configuration for the memory
uber/zap log library as one of the most popular and extensible.v2 API interfaces moved to the https://github.com/roadrunner-server/api repository. Except for logger (structure), all plugins depend only on the interfaces and don't import each other.GRPC plugin now is able to work with gzipped payloads. FR (reporter @hetao29)SQS plugin now detects EC2 env and uses AWS credentials instead of the static provider. FR (reporter @paulermo)Jobs plugin now acknowledges responses with incorrectly formed responses to prevent the infinity loop (with the error message in the logs). BUG (reporter @sergey-telpuk)protoc updated to the version v3.19.2.Nothing published for this version
Nothing published for this version
Nothing published for this version
✏️ Add the Reset method to the static pool interface. Should be used to reset the pool instead of destroying and replacing it.
Reset method to the static pool interface. Should be used to reset the pool instead of destroying and replacing it.🐛 Fix: worker exited immediately after obtaining the response. BUG (reporter: @samdark).
🐛 Fix: memory leak when supervised static pool used. PR.
✏️ New internal message bus. Available globally. Supports wildcard subscriptions (for example: http.* will subscribe you to the all events coming from
http.* will subscribe you to the all events coming from the http plugin). The subscriptions can be made from any RR plugin to any RR plugin.http:
address: 127.0.0.1:15389
middleware: [ "new_relic" ] <------- NEW
new_relic: <---------- NEW
app_name: "app"
license_key: "key"
pool:
num_workers: 10
allocate_timeout: 60s
destroy_timeout: 60s
License key and application name could be set via environment variables: (leave app_name and license_key empty)
NEW_RELIC_LICENSE_KEY.NEW_RELIC_APP_NAME.To set the New Relic attributes, the PHP worker should send headers values withing the rr_newrelic header key.
Attributes should be separated by the :, for example foo:bar, where foo is a key and bar is a value. New Relic
attributes sent from the worker will not appear in the HTTP response, they will be sent directly to the New Relic.
To see the sample of the PHP library, see the @arku31 implementation: https://github.com/arku31/roadrunner-newrelic
The special key which PHP may set to overwrite the transaction name is: transaction_name. For
example: transaction_name:foo means: set transaction name as foo. By default, RequestURI is used as the
transaction name.
$resp = new \Nyholm\Psr7\Response();
$rrNewRelic = [
'shopId:1', //custom data
'auth:password', //custom data
'transaction_name:test_transaction' //name - special key to override the name. By default it will use requestUri.
];
$resp = $resp->withHeader('rr_newrelic', $rrNewRelic);
TCP. The TCP plugin is used to handle raw TCP payload with a bi-directional protocol between the RR server and PHP worker.PHP client library: https://github.com/spiral/roadrunner-tcp
Configuration:
rpc:
listen: tcp://127.0.0.1:6001
server:
command: "php ../../psr-worker-tcp-cont.php"
tcp:
servers:
server1:
addr: 127.0.0.1:7778
delimiter: "\r\n"
server2:
addr: 127.0.0.1:8811
read_buf_size: 10
server3:
addr: 127.0.0.1:8812
delimiter: "\r\n"
read_buf_size: 1
pool:
num_workers: 5
max_jobs: 0
allocate_timeout: 60s
destroy_timeout: 60s
http_metrics.http:
address: 127.0.0.1:15389
middleware: [ "http_metrics" ] <------- NEW
pool:
num_workers: 10
allocate_timeout: 60s
destroy_timeout: 60s
All old and new http metrics will be available after the middleware is activated. Be careful, this middleware may slow down your requests. New metrics:
- `rr_http_requests_queue_sum` - number of queued requests.
- `rr_http_no_free_workers_total` - number of the occurrences of the `NoFreeWorkers` errors.
fileserver:
# File server address
#
# Error on empty
address: 127.0.0.1:10101
# Etag calculation. Request body CRC32.
#
# Default: false
calculate_etag: true
# Weak etag calculation
#
# Default: false
weak: false
# Enable body streaming for the files more than 4KB
#
# Default: false
stream_request_body: true
serve:
# HTTP prefix
#
# Error on empty
- prefix: "/foo"
# Directory to serve
#
# Default: "."
root: "../../../tests"
# When set to true, the server tries minimizing CPU usage by caching compressed files
#
# Default: false
compress: false
# Expiration duration for inactive file handlers. Units: seconds.
#
# Default: 10, use a negative value to disable it.
cache_duration: 10
# The value for the Cache-Control HTTP-header. Units: seconds
#
# Default: 10 seconds
max_age: 10
# Enable range requests
# https://developer.mozilla.org/en-US/docs/Web/HTTP/Range_requests
#
# Default: false
bytes_range: true
- prefix: "/foo/bar"
root: "../../../tests"
compress: false
cache_duration: 10s
max_age: 10
bytes_range: true
on_init option for the server plugin. on_init code executed before the regular command and can be used to warm up the application for example. Failed on_init command doesn't affect the main command, so, the RR will continue to run. Thanks (@OO00O0O)Config:
# Application server settings (docs: https://roadrunner.dev/docs/php-worker)
server:
on_init: <----------- NEW
# Command to execute before the main server's command
#
# This option is required if using on_init
command: "any php or script here"
# Script execute timeout
#
# Default: 60s [60m, 60h], if used w/o units its means - NANOSECONDS.
exec_timeout: 20s
# Environment variables for the worker processes.
#
# Default: <empty map>
env:
- SOME_KEY: "SOME_VALUE"
- SOME_KEY2: "SOME_VALUE2"
# ..REGULAR SERVER OPTIONS...
# GRPC service configuration
grpc:
# Proto files to use
#
# This option is required. At least one proto file must be specified.
proto:
- "first.proto"
- "second.proto"
## ... OTHER REGULAR GRPC OPTIONS ...
allow configuration option for the http.uploads and multipart requests. The new option allows you to filter upload extensions knowing only allowed. Now, there is no need to have a looong list with all possible extensions to forbid. FR (Thanks @rjd22)
http.uploads.forbid has a higher priority, so, if you have duplicates in the http.uploads.allow and http.uploads.forbid the duplicated extension will be forbidden.
Config:http:
address: 127.0.0.1:18903
max_request_size: 1024
middleware: ["pluginMiddleware", "pluginMiddleware2"]
uploads:
forbid: [".php", ".exe", ".bat"]
allow: [".html", ".aaa" ] <------------- NEW
trusted_subnets:
[
"10.0.0.0/8",
"127.0.0.0/8",
"172.16.0.0/12",
"192.168.0.0/16",
"::1/128",
"fc00::/7",
"fe80::/10",
]
pool:
num_workers: 2
max_jobs: 0
allocate_timeout: 60s
destroy_timeout: 60s
✏️ Beanstalk queue reject stop RPC calls if there are jobs in the priority queue associated with the requested pipeline.
✏️ Startup message when the RR has started.
pool.debug mode.v2.6.0v2.6.0v1.0.11v1.0.8v3.2.4v1.11.1Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Full Changelog: https://github.com/spiral/roadrunner/compare/v2.5.2...v2.5.3
Full Changelog: https://github.com/spiral/roadrunner/compare/v2.5.2...v2.5.3
Full Changelog: https://github.com/spiral/roadrunner/compare/v2.5.1...v2.5.2
Full Changelog: https://github.com/spiral/roadrunner/compare/v2.5.1...v2.5.2
🐛 Fix: worker's channel deadlock when using TTL. Thanks @dstrop
🔨 Some drivers now use a new config key to handle local configuration. Involved plugins and drivers:
config key to handle local configuration. Involved plugins and drivers:broadcast, kvmemory, redis, memcached, boltdb.broadcast:
default:
driver: memory
interval: 1
broadcast:
default:
driver: memory
config: {} <--------------- NEW
kv:
memory-rr:
driver: memory
config: <--------------- NEW
interval: 1
kv:
memcached-rr:
driver: memcached
config: <--------------- NEW
addr:
- "127.0.0.1:11211"
broadcast:
default:
driver: redis
config: <------------------ NEW
addrs:
- "127.0.0.1:6379"
ssl:
# Host and port to listen on (eg.: `127.0.0.1:443`).
#
# Default: ":443"
address: "127.0.0.1:443"
# Use ACME certificates provider (Let's encrypt)
acme:
# Directory to use as a certificate/pk, account info storage
#
# Optional. Default: rr_cache
certs_dir: rr_le_certs
# User email
#
# Used to create LE account. Mandatory. Error on empty.
email: you-email-here@email
# Alternate port for the http challenge. Challenge traffic should be redirected to this port if overridden.
#
# Optional. Default: 80
alt_http_port: 80,
# Alternate port for the tls-alpn-01 challenge. Challenge traffic should be redirected to this port if overridden.
#
# Optional. Default: 443.
alt_tlsalpn_port: 443,
# Challenge types
#
# Optional. Default: http-01. Possible values: http-01, tlsalpn-01
challenge_type: http-01
# Use production or staging endpoints. NOTE, try to use the staging endpoint (`use_production_endpoint`: `false`) to make sure, that everything works correctly.
#
# Optional, but for production should be set to true. Default: false
use_production_endpoint: true
# List of your domains to obtain certificates
#
# Mandatory. Error on empty.
domains: [
"your-cool-domain.here",
"your-second-domain.here"
]
logs plugin to configure the line ending. By default, used \n.New option:
# Logs plugin settings
logs:
(....)
# Line ending
#
# Default: "\n".
line_ending: "\n"
Info log level.http:
address: 127.0.0.1:55555
max_request_size: 1024
access_logs: true <-------- Access Logs ON/OFF
middleware: []
pool:
num_workers: 2
max_jobs: 0
allocate_timeout: 60s
destroy_timeout: 60s
X-Sendfile header.
Middleware reads the file in 10MB chunks. So, for example for the 5Gb file, only 10MB of RSS will be used. If the file size is smaller than 10MB, the middleware fits the buffer to the file size.http:
address: 127.0.0.1:44444
max_request_size: 1024
middleware: ["sendfile"] <----- NEW MIDDLEWARE
pool:
num_workers: 2
max_jobs: 0
allocate_timeout: 60s
destroy_timeout: 60s
server plugin:service:
some_service_1:
command: "php test_files/loop_env.php"
process_num: 1
exec_timeout: 5s # s,m,h (seconds, minutes, hours)
remain_after_exit: true
env: <----------------- NEW
foo: "BAR"
restart_sec: 1
command configuration key:server:
command: "./script.sh OR sh script.sh" <--- UPDATED
relay: "pipes"
relay_timeout: "20s"
The script should start a worker as the last command. For the pipes, scripts should not contain programs, which can close stdin, stdout or stderr.
nats:
addr: "demo.nats.io"
jobs:
num_pollers: 10
pipeline_size: 100000
pool:
num_workers: 10
max_jobs: 0
allocate_timeout: 60s
destroy_timeout: 60s
pipelines:
test-1:
driver: nats
prefetch: 100
subject: "default"
stream: "foo"
deliver_new: "true"
rate_limit: 100
delete_stream_on_stop: false
delete_after_ack: false
priority: 2
consume: [ "test-1" ]
Driver uses NATS JetStream API and is not compatible with non-js API.
✏️ Response API for the NATS, RabbitMQ, SQS and Beanstalk drivers. This means, that you'll be able to respond to a specified in the response queue. Limitations:
Job type. There are no limitations for the responses into the other queues (tubes, subjects).boltdb-jobs connection left open after RPC close command.beanstalk connection and release associated resources after pipeline stopped.reset.v2.5.0v2.5.0v1.0.10v1.0.6v3.2.3Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
🐛 Fix: bug with the not-idempotent call to the attributes.Init.
attributes.Init.🔨 Pool, worker interfaces: payload now passed and returned by the pointer.
Jobs plugin with pluggable drivers. Now you can allocate/destroy pipelines in the runtime. Drivers included in the initial release: RabbitMQ (0-9-1), SQS v2, beanstalk, memory and local queue powered by the boltdb. PRtcp|http(s)|empty [::]:port, tcp|http(s)|empty [::1]:port, tcp|http(s)|empty :// [0:0:0:0:0:0:0:1]:port) for RPC, HTTP and other plugins. RFCdebug pool config key to the .rr.yaml configuration reference.v3.2.1v1.0.9v1.0.4Nothing published for this version
Nothing published for this version
Nothing published for this version
🐛 Fix: Bug with TTL incorrectly handled by the worker PR
✏️ Rework broadcast plugin. Add architecture diagrams to the doc folder. PR
broadcast plugin. Add architecture diagrams to the doc folder. PRClear method to the KV plugin RPC. PRexec_ttl was used and TTL limit reached PRboltdb storage: Boom, Boom, Boomorigin check Bugv3.1.4v1.0.8Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
✏️ Meet the new plugin websockets with broadcast capabilities that replace old broadcast and broadcast-ws plugins. It can handle hundreds of thousands
websockets with broadcast capabilities that replace old broadcast and broadcast-ws plugins. It can handle hundreds of thousands of websocket connections simultaneously (~300k messages per second with 1k connected clients, in-memory bus on 2CPU cores and 1GB of RAM) Issuewebsockets and kv RPC calls under the hood. IssuePHP <-> GO RPC calls. It can perform more than 10% better than previously used JSON encoding/decoding for the RPC calls (especially for the KV plugin calls with a lot of Items).On top of that we released some new improvements:
latest docker image tag supported now (but we strongly recommend using a versioned tag (like 0.2.3) instead)http config section: internal_error_code to override default (500) internal error code. Issueserver.command and find errors related to the wrong path to a PHP file, or .ph, .sh scripts. Issueinformer.Workers worked incorrectly: BugSoftJob error: Bugfcgi server leads to the NPE: Bugv3.1.3v1.0.2Nothing published for this version
Nothing published for this version
Nothing published for this version
🐛 Fix: revert static plugin. It stays as a separate plugin on the main route (/) and supports all the previously announced features.
/) and supports all the previously announced features.build and other old targets from the Makefile.✏️ Reworked static plugin. Now, it does not affect the performance of the main route and persist on the separate file server (within the http plugin).
static plugin. Now, it does not affect the performance of the main route and persist on the separate
file server (within the http plugin). Looong awaited feature: Etag (+ weak Etags) as well with the If-Mach
, If-None-Match, If-Range, Last-Modified and If-Modified-Since tags supported. Static plugin has a bunch of new options such as: allow, calculate_etag, weak and pattern.
always was deleted from the plugin.Docs: static
informer.List implementation. Now it returns a list with the all available plugins in the runtime.debug mode with the exec_ttl supervisor option.🐛 Fix: issue with endure provided wrong logger interface implementation.
endure provided wrong logger interface implementation.✏️ New service plugin. Docs: link
service plugin. Docs: linkkv plugin with boltdb, in-memory, memcached and redis drivers. PHP part will be available a little bit later.Named interface implemented.rr reset.One column per quarter.