Closed smklein closed 2 weeks ago
Other notes: This was on Helios, after running cargo nt -p omicron-omdb -- test_omdb_success_cases
in a loop for almost 100 iterations.
It looks like the code that does the timestamp redactions attempts to preserve the width (in characters) of the timestamp: https://github.com/oxidecomputer/omicron/blob/eb54c1c1c013936e425c1f291af07a67f4c1c845/test-utils/src/dev/test_cmds.rs#L175-L177
It sounds like most of the time, the width matches what's in the expected output file. In this case, it was 3 characters shorter. In the past I've seen that some ways of formatting timestamps implicitly vary the precision based on the value -- e.g., if the milliseconds happens to be exactly 0, then they get left out.
On a regular run I saw this value:
warning: unknown background task: "metrics_producer_gc" (don't know how to interpret details: Object {"expiration": String("2024-09-03T17:29:27.850928397Z"), "pruned": Array []})
So (1) we're doing the string formatting on the server, and (2) my guess is that the nanoseconds component was 0 in the case we saw and the serialization just left it out.
On branch
main
commit 5bf5f09545b7b74f809c099db5c0b315ea06f9be, I saw this test failure:I was actually trying to reproduce https://github.com/oxidecomputer/omicron/issues/6505 locally, but this looks like a different error. Notably, I'm seeing differences in the whitespace between
REDACTED...TIMESTAMP
.