Fixed data race in Collect() #1140
Conversation
| return | ||
| } | ||
|
|
||
| snapshot := promAsyncMetricCache |
There was a problem hiding this comment.
I'd prefer to have a longer lock rather than duplicating metric cache that might be huge on high loaded systems
There was a problem hiding this comment.
It's not a deep copy; just a pointer swap (O(1), which is 8/4 bytes depending on OS). The map data stays in place, we just change which pointer the global variable holds. After processing, the old map gets garbage collected. New writes go to the fresh map and get picked up on the next scrape. There shouldn't be any data loss because prometheus is scraping periodically
There was a problem hiding this comment.
Also, I noticed promAsyncMetricCache is a package-level global rather than a field on the PrometheusWriter struct. Is there a specific reason for that design? Just curious if multiple writer instances are expected to share the cache intentionally
2 participants
Fixes #1136.
Will add the test as part of the PR to add tests to prometheus.go