feat(taskworker): Update Taskbroker / Taskworker Client to Support Push Mode

[george-sentry]

Linear

Completes STREAM-866

Description

Currently, taskworkers pull tasks from taskbrokers via RPC. This approach works, but has some drawbacks. Therefore, we want taskbrokers to push tasks to taskworkers instead. Read this page on Notion for more information.

This PR updates taskbroker-client package to version 0.1.8, which adds the ability to select between "pull" and "push" delivery modes. It also updates the run_taskworker command to accept extra arguments for configuring the taskworker in push mode...

• --push-mode is a boolean that turns push mode on or off (default off)
• --taskworker-rpc-port specifies the port the taskworker gRPC server should listen on (default is 50052 since the taskbroker runs on 50051, and in self or local hosted environments, brokers and workers may run on the same machine)

[linear-code]

STREAM-866 Update Sentry Taskworker

[evanh]

Nice!

[cursor]

Inconsistent sample_rate handling in new gauge method

Low Severity

The new gauge method uses sample_rate: float = settings.SENTRY_METRICS_SAMPLE_RATE as its default, while every other method in SentryMetricsBackend (incr, distribution, timer) consistently uses sample_rate: float | None = None with a runtime if sample_rate is None: check. This means gauge evaluates the setting at class-load time (freezing the value), whereas all sibling methods resolve it at call time. If the abstract MetricsBackend from taskbroker_client passes None for sample_rate, this method won't handle it correctly.

 

^{Reviewed by Cursor Bugbot for commit 36cb982. Configure here.}

[github-actions]

🚨 Warning: This pull request contains Frontend and Backend changes!

It's discouraged to make changes to Sentry's Frontend and Backend in a single pull request. The Frontend and Backend are not atomically deployed. If the changes are interdependent of each other, they must be separated into two pull requests and be made forward or backwards compatible, such that the Backend or Frontend can be safely deployed independently.

Have questions? Please ask in the #discuss-dev-infra channel.

[sentry]

🚧 Skipped: PR exceeds review size limit.

Please split into smaller PRs and re-run.
_{Reference ID: 13096878}

[george-sentry]

I have no idea what just happened, sorry 🙃

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

There are 3 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit ceedb2c. Configure here.}

[cursor]

Unhandled exceptions skip exit marker, causing CI timeout

Medium Severity

The main() function in bootstrap-snuba.py has no top-level exception handler. If an unhandled exception occurs before SNUBA_EXIT.write_text(str(rc)) runs (e.g., int(workers_str) raising ValueError for a non-numeric XDIST_WORKERS, or ThreadPoolExecutor(max_workers=0) if workers is zero), the /tmp/snuba-bootstrap-exit marker file is never written. The downstream "Wait for Snuba bootstrap" step polls this file in a loop and would silently hang for 600 seconds before timing out, making the root cause very hard to diagnose.

 

^{Reviewed by Cursor Bugbot for commit ceedb2c. Configure here.}

[cursor]

AST scope misses class-level parametrize variable references

Low Severity

The scope dict is built only from ast.iter_child_nodes(tree), which yields module-level assignments. Variables defined inside class bodies (e.g., CASES = [...] on a test class) are never captured. When ast.walk later visits parametrized methods on those classes, the variable lookup in _resolve fails and the decorator silently falls back to a count of 1 instead of the actual parameter count. This causes systematic undercounting of tests using class-level parametrize variables.

 

^{Reviewed by Cursor Bugbot for commit ceedb2c. Configure here.}