Trevor Hickey

buildfarm discussion: https://github.com/bazelbuild/bazel-buildfarm/issues/981 bazel discussion: https://github.com/bazelbuild/bazel/issues/8830#issuecomment-1004512048

[refactor] move functionality out of getActionResult

2

This change is part of https://github.com/bazelbuild/bazel-buildfarm/pull/943, but separating it here to make smaller diffs. I'd like `ShardInstance` to implement its own version of `getActionResult`, and that can be done more...

The increased disk activity from cache expirations may cause performance to worsen for ongoing executions. With this config enabled, workers can stop taking on work when they are expiring data,...

### Summary: Grpc timeouts when writing `QueuedOperations` to the CAS result in prequeue congestion which affects build performance. To resolve this, we expose the grpc timeout, polling, and concurrency mechanisms...

### Summary: Workers register themselves to the backplane so that other workers can find and communicate with them. Registering to the backplane has traditionally meant that a worker is a...

[configuration] organize prequeue and operation queue proto data

2

This organizes the config files by composing some of the data. The prequeue and operation queue are configured by name and max_depth. The existing configuration values are moved into a...

Fix expensive map size calculation via redisson

2

### Problem: Calculating size of `JedisMap` is too expensive (see issue: (https://github.com/bazelbuild/bazel-buildfarm/issues/762). Below is an explanation of the problem, and why the solution was chosen. Our map data is laid...

There appears to be two potential infinite loops during cgroups cleanup: In the controller: `while (exists) {...}` In the group: `while (!killAllProcs(controllerName)) ` We leave the implementation as-is, but ensure...

[metrics] add prometheus metrics for instance action cache

1

Buildfarm's action cache is multi-layered. There is the distributed redis implementation found in the backplane. We are recording the size for that. However, there is also a local readthrough cache...

fix static analysis warning and enable it for future PRs.