Infrastructure Failure Detection

GitAuto detects transient infrastructure failures in CI logs - segfaults, network timeouts, connection resets, out-of-memory crashes - and retries CI instead of trying to fix code. These failures are infrastructure flakiness, not code bugs.

Why This Exists

A segfault in Node.js, a network timeout during npm install, or a connection reset during a database test is not a code bug - it is infrastructure flakiness. Without detection, the agent reads the error log and tries to "fix" the code, making changes that have nothing to do with the actual problem. A simple CI re-run would pass, but the agent wastes iterations and commits on phantom bugs.

How It Works

GitAuto scans CI log output for known infrastructure failure patterns: segmentation faults, SIGKILL/SIGTERM signals, network timeout errors, DNS resolution failures, connection reset messages, and OOM killer output. When a match is found, instead of passing the log to the model for code analysis, GitAuto triggers a CI re-run directly. The re-run count is tracked separately (MAX_INFRA_RETRIES=3) to prevent infinite retries on persistently broken infrastructure.

Related Features

• Duplicate Error Hashing - detects repeated code errors after infrastructure failures are filtered out