Taskcluster

Automated Package Updates

<details>
<summary>8 Dependabot updates</summary>

* build(deps): bump the node-deps group across 1 directory with 17 updates (7664993bf)
* build(deps): bump hawk in /clients/client in the client-node-deps group (676d3a034)
* build(deps): bump the client-web-node-deps group (1724b094a)
* build(deps): bump taskcluster in /taskcluster in the python-deps group (435162d2c)
* build(deps): bump dependabot/fetch-metadata in the gh-actions-deps group (7e8f35dc2)
* build(deps): bump taskcluster-taskgraph in /taskcluster (1d2a4cae0)
* build(deps): bump the go-deps group with 5 updates (45dbb1230)
* build(deps): bump ejs from 3.1.9 to 3.1.10 (d320208b2)

</details>

WORKER-DEPLOYERS

▶ [patch] [6983](https://github.com/taskcluster/taskcluster/issues/6983)
AWS provider correctly detects `InvalidInstanceID.NotFound` error and marks worker as stopped.

USERS

▶ [patch] [6987](https://github.com/taskcluster/taskcluster/issues/6987)
Generic Worker now checks if a graceful termination was requested from worker runner _before_ calling `queue.claimWork()`.

This helps fix a race condition where a preemption occurs right after Generic Worker starts up, but before the graceful termination handler to abort the task has been initialized.

DEVELOPERS

▶ [patch] [6984](https://github.com/taskcluster/taskcluster/issues/6984)
Github auto-cancel gracefully ignores missing task groups and doesn't log errors in github comments.
This can happen when decision task failed on previous runs.

▶ [patch] [6761](https://github.com/taskcluster/taskcluster/issues/6761)
Switching from `googleapis` package to a smaller `googleapis/*` libraries to reduce startup time and avoid loading APIs we don't use.

WORKER-DEPLOYERS

▶ [patch] [6972](https://github.com/taskcluster/taskcluster/issues/6972)
Generic Worker now uploads task payload artifacts in parallel to decrease graceful termination time in the event of a spot termination.

The `insecure` engine no longer performs a file copy command as the task user before the artifact upload process happens to help speed up the process.

Generic Worker (posix only) now tries to put an exclusive file lock on artifacts before upload to prevent the file from being written to by any other process. This is done in lieu of copying the file to a temporary location which was achieving the same thing. If putting the lock on the file fails, Generic Worker will fallback to copying the file.

USERS

▶ [patch] [6972](https://github.com/taskcluster/taskcluster/issues/6972)
Don't compress `.npz` artifacts by default in Generic Worker.

Automated Package Updates

<details>
<summary>2 Dependabot updates</summary>

* build(deps): bump aiohttp from 3.9.2 to 3.9.4 in /taskcluster (c4737ab9b)
* build(deps): bump golang.org/x/net from 0.22.0 to 0.23.0 (95aa8bf85)

</details>

GENERAL

▶ [patch]
Upgrades to Node.js v20.12.2 which is a [security release](https://nodejs.org/en/blog/vulnerability/april-2024-security-releases-2/).

Automated Package Updates

<details>
<summary>6 Dependabot updates</summary>

* build(deps): bump idna from 3.4 to 3.7 in /taskcluster (2cd73d93a)
* build(deps): bump tar from 6.2.0 to 6.2.1 in /ui (254af1652)
* build(deps): bump tar from 6.2.0 to 6.2.1 in /clients/client (f8ddfbeb7)
* build(deps): bump tar from 6.2.0 to 6.2.1 (e1ff1fb14)
* build(deps): bump tar from 6.2.0 to 6.2.1 in /clients/client-test (3b9ec5c1a)
* build(deps): bump tar from 6.2.0 to 6.2.1 in /clients/client-web (f69bea143)

</details>

WORKER-DEPLOYERS

▶ [patch] [6900](https://github.com/taskcluster/taskcluster/issues/6900)
Worker Runner on Azure no longer sends a `graceful-termination` message if the scheduled event type is `Freeze`. It will continue to send the message for all other event types: `Reboot`, `Redeploy`, `Preempt`, and `Terminate`.

USERS

▶ [patch] [6957](https://github.com/taskcluster/taskcluster/issues/6957)
Improves performance of the `findTasksAtIndex` call (introduced in 6915)
that returns multiple tasks for a given list of namespaces and indexes.

GENERAL

▶ [patch] [6958](https://github.com/taskcluster/taskcluster/issues/6958)
Worker Manager now only applies GCP disk labels to `PERSISTENT` disk types.

This fixes an issue in v64.2.2 where `initializeParams.labels` was being set on all disk types and caused GCP to error on local SSDs (`SCRATCH` type disks).