Remove HardLimitingWorker by @SchrosCat2013 fixing #6424

[pavelbrylov]

What type of PR is this?

• [X ] Bug Fix

Description

There is an issue with adhoc queries running longer than minute, there a lot of details in #6424 (comment)
Author of investigation and changes is @SchrosCat2013, I just adopted changes in his fork to the lates master in a hope of not having to maintain own fork :)

How is this tested?

• [ X ] Manually

We patched and built own image, this solves the issue and same long running query is now polled until it completes as opposed to stop polling in 1 minute.

Related Tickets & Documents

#6424

[pavelbrylov]

@justinclift if you could help prioritize this - would be great :) Thanks!

[greptile-apps]

Greptile Summary

This PR removes Redash's custom HardLimitingWorker, delegating work-horse monitoring and timeout enforcement to RQ while retaining StatsD instrumentation and the custom cancellable job type.

• Simplifies RedashWorker inheritance.
• Removes custom timeout, heartbeat, process-exit, and cancellation monitoring.
• Leaves running-query cancellation without the signal path needed to interrupt query execution.

Confidence Score: 4/5

This PR is not safe to merge until running-query cancellation still interrupts the active work horse.

The query cancellation endpoint now only updates job state; removing the custom monitor eliminates the path that observes that state and signals the process executing the database query.

Files Needing Attention: redash/tasks/worker.py

Important Files Changed

Filename	Overview
redash/tasks/worker.py	Removes the custom work-horse monitor and hard-limit implementation, but also drops the signal-based handling required by Redash's running-job cancellation flow.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[DELETE running query job] --> B[CancellableJob.cancel]
  B --> C[Store cancelled metadata]
  C --> D{Worker monitor}
  D -->|Before this PR| E[Refresh cancellation state]
  E --> F[Send SIGINT to work horse]
  F --> G[Query runner cancels database operation]
  D -->|After this PR| H[Inherited RQ monitoring]
  H --> I[No Redash cancellation signal]
  I --> J[Query process may continue running]

Loading

_{Reviews (1): Last reviewed commit: "Merge branch 'master' into master" | Re-trigger Greptile}

[greptile-apps]

Running Cancellation No Longer Interrupts

Removing HardLimitingWorker also removes the monitor path that refreshed Redash's cancellation metadata and sent SIGINT to a running work horse. When the query-job DELETE endpoint calls CancellableJob.cancel(), the job is marked cancelled in Redis, but the executing query process is not interrupted. Its query-runner cancellation handler is therefore never invoked, and the query can continue consuming worker and database capacity.

Knowledge Base Used: Background jobs and scheduling