redpanda
redpanda copied to clipboard
redpanda-data
Cloud retention stress test
Cover letter
Test cloud retention under load. The consume load comes in two flavours:
- an ordinary fast consumer that repeatedly consumes the tail few segments
- a slow consumer that can't keep up with retention and has to start over when retention overcomes it
Backport Required
- [x] v22.3.x
UX changes
none
Release notes
none
Is the stress test itself that it only does historical reads. Or is the stress test mixed.
Wanting to understand cover letter
@emaxerrno The reads are mainly historical - the fast consumer manages to read a few segments before being overtaken, but still can't keep up with the produce rate.
debug build failed after all tests successfully finished:
Terminated with signal SIGKILL (from the agent because the job was canceled). Exit status -1
Failed to upload one of the artifacts maybe?
restarted
https://buildkite.com/redpanda/redpanda/builds/18620#01847da2-d436-4c8a-a501-fff95aa8184d
unknown location(0): [4;31;49mfatal error: in "read_from_ntp_max_bytes": std::runtime_error: ntp {kafka/my_topic_0/0}: log offset -9223372036854775808 is outside the translation range (starting at 0)[0;39;49m
hmm...