NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #4526 most downloaded on PyPI
Python client for Taskcluster
Latest stable
116.0.0
Last release 12 days ago
10 Nov 2026
Ships on a steady schedule
a new release about every 2 weeks
Nearly every release is documented
notes for 60 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
12 years old
502 releases · first in 2014
Drag across the ruler to compare, or tap a release to open it.
▶ [patch] Upgrades goreleaser tool to latest version to fix broken release. Release v95.1.2 did not publish, so its changelog will be effective for v9
▶ [patch] Upgrades goreleaser tool to latest version to fix broken release. Release v95.1.2 did not publish, so its changelog will be effective for v95.1.3.
▶ [patch] bug 2006698 Fixes worker-manager killing zombie workers when queue data is missing. For long running tasks that took longer than queueInacti
▶ [patch] bug 2006698 Fixes worker-manager killing zombie workers when queue data is missing. For long running tasks that took longer than queueInactivityTimeout to finish was a high risk of getting claim-expired because w-m incorrectly assumed worker is not doing anything (missing queue information)
Your coding agent can read these notes before it upgrades. Set up the MCP server →
▶ [minor] Queue service allows changing the priority of unresolved tasks. Two new endpoints are being introduced:
▶ [minor] Queue service allows changing the priority of unresolved tasks. Two new endpoints are being introduced:
This is the implementation of the RFC190
▶ [patch] Notify Service: upgrades to use @aws-sdk/client-sesv2 with the latest version of nodemailer to resolve some security vulnerabilities.
▶ [patch]
Notify Service: upgrades to use @aws-sdk/client-sesv2 with the latest version of nodemailer to resolve some security vulnerabilities.
▶ [patch] Upgrades CI Decision task to use Taskgraph v18.0.3.
▶ [patch] Upgrades to Node.js v24.12.0.
▶ [patch] #8176 Worker-manager ensures workers are spawned for single launch config worker pools when adjusted weight might drop below zero with remaining capacity.
▶ [patch] Changed the garbage collector to clean caches after docker resources when d2g is enabled
▶ [patch] Switch back web-server queues to classic, as those are only used for short-lived task group updates in the UI and don't require same durability and Raft consensus algorithm.
<details> <summary>9 Dependabot updates</summary>
</details>
▶ [patch] #8083 Generic Worker (windows): adds retries to the win32 LoadUserProfile call to help prevent The device is not ready worker errors.
▶ [patch] #8083
Generic Worker (windows): adds retries to the win32 LoadUserProfile call to help prevent The device is not ready worker errors.
▶ [patch] Generic Worker (with worker-runner): properly unregisters a worker when it exits due to internal error or non-current deployment ID. Followup to #8165.
▶ [patch] #8023 Generic Worker: limits concurrent artifact uploads to 10 at a time. Followup to #8032.
▶ [patch] #8153 Generic Worker (d2g): anonymous volumes from the task container are no longer removed at the end of a task run in order to resolve the
▶ [patch] #8153 Generic Worker (d2g): anonymous volumes from the task container are no longer removed at the end of a task run in order to resolve the task sooner. The garbage collector was updated to take care of these instead.
▶ [patch]
Generic Worker: moves the AbortFeature to be the first to initialize and start up for each task. This change allows the worker to be ready for spot terminations as early as possible, so that it doesn't get preempted while it's setting up other task features, not knowing the cloud provider wants the worker to shutdown.
This will help tasks resolve properly as worker shutdown in the case where a spot termination request comes in as soon as the worker starts up. Beforehand, when this situation would happen, the task would commonly resolve as claim expired because the worker didn't have enough time to react to the preemption notice.
▶ [patch] Worker-runner will now properly unregister a worker when it exits because of an idle timeout
▶ [MAJOR] #8074 Pulse consumer expects core Taskcluster classic RabbitMQ queues to be deleted before being auto-recreated as quorum queues.
▶ [MAJOR] #8074 Pulse consumer expects core Taskcluster classic RabbitMQ queues to be deleted before being auto-recreated as quorum queues.
The API reference documentation now includes code examples for every endpoint in multiple languages:
▶ [minor] #8049
The API reference documentation now includes code examples for every endpoint in multiple languages:
Each example demonstrates how to:
Examples are generated dynamically in the browser from API metadata, keeping the documentation lean while providing comprehensive, up-to-date code samples. Examples include syntax highlighting and can be copied to clipboard directly from the documentation.
▶ [patch] #8074 Fixes pulse consumer issue where services would assert the queue exists as a quorum queue and wouldn't fall back to classic type as a backwards compatibility followup solution to #8156.
edit: This should've been a breaking change due to the fact that old, classic queues cannot be converted to quorum type in place. You must delete the…
▶ [minor] #8156 Pulse library declares non-ephemeral core Taskcluster queues as quorum queues to prepare for upgrading to RabbitMQ v4+.
edit: This should've been a breaking change due to the fact that old, classic queues cannot be converted to quorum type in place. You must delete the old classic queue and let the pulse consumer recreate it as a quorum queue. Please upgrade to at least v95.0.0.
build(deps): bump jws from 3.2.2 to 3.2.3
<details> <summary>1 Dependabot updates</summary>
</details>
▶ [patch] Client (python): upgrades many dependencies to latest minor/patch releases using uv lock --upgrade.
▶ [patch]
Client (python): upgrades many dependencies to latest minor/patch releases using uv lock --upgrade.
▶ [patch] Upgrades to Node.js v24.11.1 and rust v1.91.1. Additionally upgrades yarn to 4.12.0.
Replaces backoff crate with backon due to https://rustsec.org/advisories/RUSTSEC-2025-0012
▶ [patch] Upgrades to go1.25.5 [SECURITY].
See more here.
▶ [patch] #8115
Generic Worker (windows): reverts #8030 to use CREATE_NEW_CONSOLE over CREATE_NO_WINDOW so that child processes can call AllocConsole() to create new consoles.
▶ [minor] #8093 Github webhook endpoint returns 200 instead of 400 for unsupported events. 200 means we received and processed webhook, even if we don't actually support such event at the moment. 400 is only for validation issues.
<details> <summary>8 Dependabot updates</summary>
</details>
▶ [patch] Fixes errors handling for upgraded googleapis packages. Instance creation errors were sent differently, which didn't allow to log some provi
▶ [patch] Fixes errors handling for upgraded googleapis packages. Instance creation errors were sent differently, which didn't allow to log some provision exceptions.
▶ [patch] Azure provider logs additional error details for failed ARM deployments.
▶ [patch] Azure provider logs additional error details for failed ARM deployments.
▶ [MAJOR] #7765
Spaces in artifact names are now correctly preserved instead of being replaced with +
▶ [patch] #7829 Fix a bug that prevented all-resolved tasks from getting scheduled if they depended on a task that was also part of an all-completed dependency and that all-completed task was processed before the all-resolved one
▶ [patch] Increase the buffer size of websocktunnel on the worker side to make livelogs a lot faster
▶ [patch] #8121 The taskcluster CLI now encodes URL path parameters as path parametes rather than as query string parameters.
<details> <summary>2 Dependabot updates</summary>
</details>
▶ [patch] Azure provider logs additional error details for failed ARM deployments.
▶ [MAJOR] #7765
Spaces in artifact names are now correctly preserved instead of being replaced with +
▶ [MAJOR] Increase the buffer size of websocktunnel on the worker side to make livelogs a lot faster.
This is a breaking change, since the pre-v94 Websock Tunnel service will no longer work with newer (v94+) clients (workers). However, older clients (pre-v94 workers) will continue to work with the new Websock Tunnel service.
IMPORTANT: Therefore, it is imperative, that the Websock Tunnel service should be updated to v94+ before any clients (workers) are updated to v94+.
▶ [patch] #7829 Fix a bug that prevented all-resolved tasks from getting scheduled if they depended on a task that was also part of an all-completed dependency and that all-completed task was processed before the all-resolved one
▶ [patch] #8121 The taskcluster CLI now encodes URL path parameters as path parametes rather than as query string parameters.
<details> <summary>2 Dependabot updates</summary>
</details>
▶ [patch] #8102 Don't try to chown caches when using d2g as this is a long no-op anyway
▶ [patch] #8102 Don't try to chown caches when using d2g as this is a long no-op anyway
▶ [patch] Make livelogs faster when passing through websocktunnel by increasing the buffer size on the proxy
▶ [patch] Worker-manager improves handling of workers created for ARM deployment that failed validation step
▶ [patch] Worker-manager improves handling of workers created for ARM deployment that failed validation step
▶ [patch] #8083
Generic Worker (windows): adds retry logic around CreateUserProfile method to ensure the task user's profile path is created successfully before continuing on to LoadUserProfile.
<details> <summary>2 Dependabot updates</summary>
</details>
▶ [patch] Worker registration metrics exposed from all relevant pods.
▶ [patch] Worker registration metrics exposed from all relevant pods.
<details> <summary>2 Dependabot updates</summary>
</details>
▶ [patch] #8086 Worker-manager Azure ARM deployment fix for stuck worker in removal/stopping loop, better descriptions in reported errors.
▶ [patch] #8086 Worker-manager Azure ARM deployment fix for stuck worker in removal/stopping loop, better descriptions in reported errors.
▶ [patch] #8083
Generic Worker (windows): fixes intermittent issue calling LoadUserProfile win32 API when the device isn't ready.
▶ [patch] #8082 Relax GitHub webhook identifier regex to accept [bot] senders.
▶ [patch] Worker-manager web services exposes metrics now.
▶ [patch] Worker-manager web services exposes metrics now.
▶ [patch] #8079 Generic Worker: logs artifact upload status to worker logs instead of the task log, unless there was an error uploading.
▶ [minor] Worker-manager adds extra Prometheus metrics for worker registration timing: worker_manager_worker_registration_seconds - time from being re
▶ [minor]
Worker-manager adds extra Prometheus metrics for worker registration timing:
worker_manager_worker_registration_seconds - time from being requested to registered (running)
worker_manager_worker_lifetime_seconds - total lifetime of a worker (until stopped or removed)
worker_manager_worker_registration_failures_total - count of workers that were removed before register
▶ [patch] #8077 Azure provider reports ARM template deployments errors on the worker pool level. When deployment fails and one or more resources were not created, errors were hidden in operations list, which made it difficult to debug.
▶ [patch] #8075 Web-server improves graphql validation
▶ [MAJOR] #7965 GitHub service now validates webhook payloads against a schema to prevent TypeErrors from missing or malformed fields. This ensures th
▶ [MAJOR] #7965 GitHub service now validates webhook payloads against a schema to prevent TypeErrors from missing or malformed fields. This ensures that webhook handlers receive properly structured data, the webhook now returns a 400 status and does not create an unprocessable event.
The githubWebHookConsumer endpoint has been removed from client libraries as it was not intended for client use (only for GitHub webhook integrations).
▶ [minor] #8068 Client (python): adds python 3.9 support back due to Taskgraph still needing to support it.
▶ [patch] Upgrades to go1.25.4
▶ Additional change not described here: #8065.
▶ Additional change not described here: #8065.
▶ [patch] #8061 Updates the python client's uv.lock file to properly bump the taskcluster version on yarn release.
▶ [patch] Upgrades to Node.js v24.11.0
▶ [patch] Upgrades to Node.js v24.11.0
▶ [patch] Worker-scanner azure now exposes metrics to prometheus too.
▶ [patch] #8042 DB adds primary key to the tcversion table
▶ [patch] #7376 Check-in the initial code for generating Firefox Profiler profiles for task logs and task groups.
<details> <summary>1 Dependabot updates</summary>
</details>
▶ [patch] Worker Runner now doesn't allow workers to be bootstrapped with zip files containing the Zip Slip vulnerability.
▶ [patch] Upgrades to Node.js latest LTS, v24.10.0
▶ [patch] #8012
Generic Worker (windows): creates no cmd.exe window that appeared during task execution.
▶ [patch] Worker Runner now doesn't allow workers to be bootstrapped with zip files containing the Zip Slip vulnerability.
▶ [patch] Fixes documentation typos and links
▶ [patch] Replace minio with localstack for local development
▶ [minor] #7979 Azure worker pools can now provision via ARM templates in launch configs, reducing provisioning latency and letting Azure cascade reso
▶ [minor] #7979 Azure worker pools can now provision via ARM templates in launch configs, reducing provisioning latency and letting Azure cascade resource creation/cleanup automatically.
▶ [patch] #7979
Azure ARM template deployments allow keeping history for debug purposes by using workerManager.keepDeployment: true
▶ [patch] #8023 Generic Worker: limits concurrent artifact uploads to 100 at a time.
▶ [patch] #8031 GitHub service handles missing users (404 from getByUsername) when processing webhooks.
▶ [patch] bug 1990567 Python client exceptions are now picklable.
▶ [patch] Client (python): fixes release of client. v91.0.3 did not work in fixing.
▶ [patch] Client (python): fixes release of client. v91.0.3 did not work in fixing.
▶ [patch] Switches CI tasks to use uv tool and to the latest stable Debian release (trixie).
▶ [patch]
Switches CI tasks to use uv tool and to the latest stable Debian release (trixie).
▶ [patch] Upgrades to taskgraph v16.2.1
▶ [patch] Improves Azure resource deprovisioning by skipping checks on already deleted resources. Previously implemented logic was flawed in a way that same resources would be queried over and over. Which led to an increased number of cloud api calls and was likely causing some minor delays per each worker being deprovisioned
▶ [MAJOR] #8010 Python client follows redirects in artifact download for both async and sync code. This was partially fixed in the past in #4057
▶ [patch] Upgrades to go1.25.3
▶ [MAJOR] #8010 Python client follows redirects in artifact download for both async and sync code. This was partially fixed in the past in #4057
▶ [patch] Upgrades to Node.js v22.20.0, go1.25.2 [SECURITY], yarn 4.10.3, and golangci-lint 4.5.0.
▶ [patch] Upgrades to Node.js v22.20.0, go1.25.2 [SECURITY], yarn 4.10.3, and golangci-lint 4.5.0.
▶ [patch] #8004 D2G: always pulls named docker images to fetch latest tag.
<details> <summary>9 Dependabot updates</summary>
</details>
▶ [patch] bug 1990389 D2G: add stderr logging if any docker commands fail within the task feature.
▶ [patch] bug 1990389 D2G: add stderr logging if any docker commands fail within the task feature.
▶ [patch] #7974
D2G: fixes docker: invalid env file (env.list): bufio.Scanner: token too long error when providing the docker run command an --env-file that contains a line longer than 64KiB. D2G now passes the variable directly to the run command with -e <envVarName>=<envVarValue> to work around this constraint.
▶ [patch]
D2G: removes exit codes 125, 128 from payload.onExitStatus.retry array, as docker pulls now happen outside of the payload being run (inside the D2G task feature startup).
▶ [patch] #7969 D2G: fixes issue with loading the docker image artifact if the image already exists with a separate ID, causing the following output t
▶ [patch] #7969
D2G: fixes issue with loading the docker image artifact if the image already exists with a separate ID, causing the following output to be improperly parsed by D2G: The image <imageName> already exists, renaming the old one with ID <sha> to empty string.
▶ [patch] #7967 D2G: accounts for image artifacts that may contain multiple tags for the same image, previously causing a worker error: runtime error:
▶ [patch] #7967
D2G: accounts for image artifacts that may contain multiple tags for the same image, previously causing a worker error: runtime error: index out of range [1] with length 1.
▶ [patch] D2G: creates a Config struct type for D2G-specific config used in the repo.
▶ [patch]
D2G: creates a Config struct type for D2G-specific config used in the repo.
▶ [patch]
D2G: remove call out to bash shell that ran the docker run command. Directly call docker as the resulting Generic Worker payload command.
▶ [patch] #7942 New metrics are being exposed to prometheus for scraping:
▶ [patch] #7942 New metrics are being exposed to prometheus for scraping:
http_requests_total http requests per service/method/namehttp_request_duration_seconds http request duration histogramauth_success_total successful authentication attempt per clientId and schemeauth_failure_total failed authentication attempts and reasonsExisting queue metrics queue_failed_tasks, queue_exception_tasks now includes reasonResolved label
▶ [MAJOR] #7947 Generic Worker (Windows): Windows/386 platform is now unsupported and no longer built/released. Windows arm64 support has been added as a Tier-2 platform (guaranteeing to build).
▶ [patch] D2G: remove call out to bash during garbage collection of docker images
▶ Additional change not described here: #7942.
▶ [MAJOR] The dind and dockerSave features are no longer supported in docker-worker payloads (whether in docker-worker itself or in generic-worker via
▶ [MAJOR]
The dind and dockerSave features are no longer supported in docker-worker
payloads (whether in docker-worker itself or in generic-worker via D2G). Tasks
requesting those features will fail with an exception.
▶ [patch] #7605 Fixes duplicate tasks shown in claimed tasks list. This can happen when task is being reclaimed and multiple entries might still exist in queue_claimed_tasks table
▶ [patch]
D2G: don't write env vars out to Generic Worker's environment. Instead, pass them as -e <name>=<value>. Follow-up to https://github.com/taskcluster/taskcluster/pull/7945#discussion_r2348906304.
▶ [patch] #7938
D2G: pass Docker Worker environment variables to container using --env-file in docker run command.
▶ [patch] D2G: remove calls out to bash shells to load docker image and create chain of trust additional data file.
▶ [patch] #7500 Github build is marked as failed after there are no retries left, instead of staying "neutral"
<details> <summary>1 Dependabot updates</summary>
</details>
▶ [patch] Worker Runner (Windows): capture Generic Worker service exit code and exit early if the worker is rebooting, preventing a Worker Manager rem
▶ [patch]
Worker Runner (Windows): capture Generic Worker service exit code and exit early if the worker is rebooting, preventing a Worker Manager removeWorker API call.
▶ [patch] Upgrades to Node.js 22.19.0, go1.25.1, and yarn 4.9.4.
▶ [patch] Upgrades to Node.js 22.19.0, go1.25.1, and yarn 4.9.4.
▶ [patch]
Worker Runner: adds additional logging around sending graceful-termination request to worker.
Worker Runner (windows): fixes protocol pipe connection so that Generic Worker can communicate with Worker Runner. This allows graceful-termination requests to be properly sent and received, among other message types. You must include --with-worker-runner in your Generic Worker service configuration on the run subcommand.
▶ [patch] Bumps github.com/ulikunitz/xz from v0.5.14 to v0.5.15 to fix our release process. v88.1.0 was broken due to https://github.com/ulikunitz/xz/
▶ [patch]
Bumps github.com/ulikunitz/xz from v0.5.14 to v0.5.15 to fix our release process. v88.1.0 was broken due to https://github.com/ulikunitz/xz/issues/65.
Initially said No changes due to the fact that we had to fix an issue with the release for v88.0.4. The equivalent release notes for this release are
Initially said No changes due to the fact that we had to fix an issue with the release for v88.0.4. The equivalent release notes for this release are the ones from v88.0.4 minus the Dependabot upgrade.
▶ [patch] Upgrades to go1.25.0.
▶ [patch]
Worker Runner/Generic Worker (Azure): polls the metadata service for events the worker should gracefully terminate on every second (down from 15s). This frequency is recommended by Microsoft here and will hopefully reduce tasks resolving as claim-expired.
▶ [patch] Reverts taskcluster/taskcluster#7920 due to https://github.com/google/go-containerregistry/issues/2072 which broke release v88.0.4.
▶ [patch] Upgrades to Node.js v22.18.0 and go1.24.6 [SECURITY].
▶ [patch] Upgrades to Node.js v22.18.0 and go1.24.6 [SECURITY].
▶ [patch] #7903
Worker Manager (Azure): uses Standard SKU public IP addresses, as Basic will be retired on September 30, 2025. Official announcement
▶ [patch] #7914
Worker Manager (Azure): uses Static public IP allocation with the Standard SKU.
▶ [patch] #7905 Generic Worker launch agent on macOS for multiuser engine was crashing due to a bug in the launch daemon that was garbage collecting file handles while the agent was using them. This resulted in the launch agent crashing. This has been fixed by keeping file handles alive in the daemon.
<details> <summary>7 Dependabot updates</summary>
</details>
▶ [patch] UI: use @taskcluster/client-web package.
▶ [patch]
UI: use @taskcluster/client-web package.
▶ [patch] #7680 Worker Manager (AWS): terminates workers that are in the stopped state.
▶ Additional change not described here: #6854.
<details> <summary>3 Dependabot updates</summary>
</details>
▶ [patch] Fix releases of @taskcluser/client and @taskcluster/client-web by making their access public by default.
▶ [patch]
Fix releases of @taskcluser/client and @taskcluster/client-web by making their access public by default.
<details> <summary>5 Dependabot updates</summary>
</details>
▶ [MAJOR] #7854 Renames all Node packages to be in the @taskcluster scope.
▶ [MAJOR] #7854
Renames all Node packages to be in the @taskcluster scope.
taskcluster-client --> @taskcluster/clienttaskcluster-client-web --> @taskcluster/client-web▶ [patch] #7719 Expired secrets leave an audit event/record.
▶ [patch] #7719 Expired secrets leave an audit event/record.
▶ [patch] Fixes UI actions form that was broken with the last refactoring.
<details> <summary>2 Dependabot updates</summary>
</details>
▶ [patch] #7861 Generic Worker (D2G): fixes docker image caching issue when the worker loaded different images with the same image name, overwriting t
▶ [patch] #7861 Generic Worker (D2G): fixes docker image caching issue when the worker loaded different images with the same image name, overwriting the previously loaded docker image.
▶ [patch] #7863
Generic Worker (D2G): reloads cache on d2g task feature start to fix Unable to find image '<image>' locally issue due to garbage collection running between tasks and pruning all docker images.
▶ [patch] #7844 Worker Manager: bumps Azure SDKs to use latest API versions in order to deploy v6 SKUs.
https://nodejs.org/en/blog/vulnerability/july-2025-security-releases
▶ [patch] #7849 Generic Worker (D2G): caches loaded docker images.
▶ [patch] Upgrades decision task to v14.4.1, rust toolchain to 1.88.0, and golangci-lint version to 2.2.2.
▶ [patch] Upgrades to Node.js v22.17.1 [SECURITY].
https://nodejs.org/en/blog/vulnerability/july-2025-security-releases
▶ [patch] #7856
Fixes queue.taskQueueCounts() numbers for the total claimed tasks.
Due to the internal structure some task/runs might have had duplicate entries in the queue_claimed_tasks table
which led to slightly higher counts returned by queue_claimed_tasks_count(taskQueue) function.
<details> <summary>1 Dependabot updates</summary>
</details>
▶ [patch] #7833 Node deprecated url.parse in favor of URL.parse()
▶ [minor] #7840
Generic Worker: adds status subcommand to output task ID if a task is currently being executed.
Example output while a task is running:
$ generic-worker status
{
"currentTaskIds": [
"fz7IO4uCTtevuLBoq8Qz3w"
]
}
Example output while worker is idle:
$ generic-worker status
{
"currentTaskIds": []
}
▶ [patch] #7842
The task-running event now includes task.tags, consistent with other lifecycle events.
▶ [patch] #7846
UI: properly handle Worker Pool config changes and enable/disable Save Worker Pool button dynamically.
▶ [patch] #7842 task-pending and task-exception pulse message now consistently include the task's tags.
▶ [patch] #7833
Node deprecated url.parse in favor of URL.parse()
▶ [patch] Upgrades to Node.js 22.17.0 and yarn 4.9.2
▶ [patch] Upgrades to Node.js 22.17.0 and yarn 4.9.2
▶ [patch] Upgrades to go1.24.5 [SECURITY].
Go release notes here.
▶ [MAJOR] Client (python): Drops python 3.8 support, adds support and tests for 3.12 and 3.13.
<details> <summary>8 Dependabot updates</summary>
</details>
▶ [patch] Generic Worker: prefer strconv.ParseUint() over strconv.Atoi() when resulting int is converted to an int type of smaller size to prevent une
▶ [patch]
Generic Worker: prefer strconv.ParseUint() over strconv.Atoi() when resulting int is converted to an int type of smaller size to prevent unexpected values.
▶ [patch] #6951 Fixes UI issue with state when tasks change. Sometimes actions could be fired against wrong task group.
▶ [patch] Generic Worker multiuser on macOS now ensures finalisers are not called on launch agent pipes while they are still in use.
▶ [patch] Generic Worker: Fix for intermittent "sendmsg: bad file descriptor" on macOS multiuser.
▶ [patch] Generic Worker: Fix for intermittent "sendmsg: bad file descriptor" on macOS multiuser.
▶ [MAJOR] #7356 Generic Worker multiuser engine on macOS now executes all task commands via a Launch Agent running in the context of the desktop sessi
▶ [MAJOR] #7356 Generic Worker multiuser engine on macOS now executes all task commands via a Launch Agent running in the context of the desktop session. This means that task commands now have full access to desktop session services, such as the clipboard.
On macOS it is essential that after updating the generic-worker binary,
before starting the worker up, that the file next-task-user.json is deleted.
This file should only be deleted when upgrading the worker, not every time the
worker runs. This will cause Generic Worker to create new task users, which is
needed for the launch agent to work. Otherwise the new Generic Worker would try
to use task users created by the old Generic Worker user, which would not work.
Note, the full path to this file can be seen in the generic worker logs, if you
are not sure where to find it on the file system.
Note, this only needs to be done once when installing the new version of Generic Worker. After that, Generic Worker will continue to mange this file itself as normal.
<details> <summary>4 Dependabot updates</summary>
</details>
▶ [patch] #7789 Generic Worker: Artifact uploads properly fail on 4xx status codes. Issue first noticed in v84.1.1
▶ [patch] #7789 Generic Worker: Artifact uploads properly fail on 4xx status codes. Issue first noticed in v84.1.1
▶ [patch] #7786 Github service comments back when taskcluster command in issue comment didn't produce any tasks
▶ [patch] bug 1559766
d2g adds localhost.localdomain to /etc/hosts
▶ [MAJOR] Generic Worker: renamed the following worker configuration options to avoid confusion:
▶ [MAJOR] Generic Worker: renamed the following worker configuration options to avoid confusion:
absoluteHighMemoryThreshold --> minAvailableMemoryBytesrelativeHighMemoryThreshold --> maxMemoryUsagePercent▶ [patch] #7498 Further UI memory optimizations for the TaskGroup page to avoid using too much memory during intensive updates.
▶ [patch] #7498 Further UI memory optimizations for the TaskGroup page to avoid using too much memory during intensive updates.
▶ [minor] #7770 Generic Worker: adds additional resource monitoring auto-abortion configuration to better fine-tune how your worker aborts running tas
▶ [minor] #7770 Generic Worker: adds additional resource monitoring auto-abortion configuration to better fine-tune how your worker aborts running task processes.
absoluteHighMemoryThreshold: The minimum amount of available memory (in bytes) required before considering task abortion. If available memory drops below this value, it may trigger an abort. Default: 524288000 (500MiB).relativeHighMemoryThreshold: The percentage of total system memory usage that, if exceeded, contributes to the decision to abort the task. Default: 90.allowedHighMemoryDurationSecs: The maximum duration (in seconds) that high memory usage conditions can persist before the task is aborted. Default: 5.Generic Worker will auto-abort a task if the total system memory used percentage is greater than relativeHighMemoryThreshold AND the available memory is less than absoluteHighMemoryThreshold for longer than allowedHighMemoryDurationSecs, unless disableOOMProtection is enabled.
▶ [patch] #7769 Generic Worker: resource monitor will print out its usage summary after aborting the task.
<details> <summary>5 Dependabot updates</summary>
</details>
▶ [patch] #7763 Helm chart forces metrics-only deployments to have replicas: 0 if prometheus.enabled is false
▶ [patch] Upgrades to Node.js v22.16.0 and go1.24.4 [SECURITY].
▶ [patch] Upgrades to Node.js v22.16.0 and go1.24.4 [SECURITY].
▶ [patch] #7498 Optimizes UI memory usage for some pages that were caching table data
▶ Additional change not described here: bug 1971309.
<details> <summary>1 Dependabot updates</summary>
</details>
▶ [patch] Modernized logger interface by replacing deprecated logger.warn with logger.warning to avoid DeprecationWarning in Python environments.
▶ [patch] #7712 Upgrades to go1.24.3 [SECURITY].
More details here.
▶ [minor] #7707
Adds Prometheus metrics support to the monitor via a new plugin. Metrics can now be registered using MonitorManager.registerMetric(), similar to log types. When enabled, each configured service and background job starts a separate server on port 9100 to expose metrics for Prometheus scraping.
Example minimal Kubernetes values.yml configuration:
prometheus:
enabled: true
prefix: tc
server:
ip: 0.0.0.0
port: 9100
If your cluster does not support the monitoring.googleapis.com/v1/PodMonitoring resource, add "podmonitoring" to .skipResourceTypes[] to prevent deployment failures.
▶ [minor] #7646 The github service will now respect github rate limits instead of hammering against it as fast as it can
▶ [patch] #7757 Fixed input validation for some queue service endpoints to return a 400 error instead of a 500 error when validation fails.
▶ [patch] #7612 Exposes worker-manager provision loop metrics to prometheus.
▶ [patch]
Modernized logger interface by replacing deprecated logger.warn with logger.warning to avoid DeprecationWarning in Python environments.
<details> <summary>10 Dependabot updates</summary>
</details>
▶ [patch] #4943 Fixes UI error "buildSignedUrl missing required credentials" shown directly after login due to missing credentials.
▶ [patch] #4943 Fixes UI error "buildSignedUrl missing required credentials" shown directly after login due to missing credentials.
▶ [patch] Upgrades to Node.js v22.15.1 (SECURITY).
▶ [patch] Upgrades to Node.js v22.15.1 (SECURITY).
More info here.
▶ [patch] bug 1967254
Generic Worker (d2g): uses info log level when uploading error artifacts if the artifact was marked optional. Previously, the log would contain [taskcluster:error] which Treeherder would identify as a failure, even if the task is successful.
▶ [patch] #7630 Generic Worker: improves artifact upload process by parallelizing the scanning of the system for file/directory artifacts from the payload.
One column per quarter.