LinAdams0096 · 8 Dec 2024, 00:27 UTC
Check the timestamps first, because my display once showed old readings beautifully while everyone blamed the network.
11 pointsThe displayed timestamp stops as well, then jumps ahead with no readings shown for the gap.
7 pointsThat leaves two possibilities in my head: missing samples or samples waiting somewhere, can you see whether the file has them?
8 pointsTime the read and file write separately. The screen alone can't tell you which waited.
16 pointsDid those timings show anything? I ask because rotation was when our report tool copied the previous file to a slow share.
11 pointsFound a copy-to-archive call in the rotation handler, it sits on the same loop as the reads and the delay matches the freeze... I hadn't realised that archive folder was remote.
12 pointsThat's promising, ours recovered once archiving stopped holding up acquisition, though we still had to show when the archive was behind.
13 pointsDoes your proposed background writer have a bounded queue and an explicit overflow policy? Moving the wait can otherwise turn a visible stall into silent memory growth, which is a pain to hand over.
11 pointsI've moved archival copies off the sampling loop and capped the pending work, failed copies stay local with an alert; a deliberately unavailable share no longer freezes the viewer, and the sampler timings stay steady.
17 pointsGood, the unavailable-share test is the bit I wanted to hear about, our first fix worked only while everything was behaving.
18 pointsThis discussion is closed to new replies after six months without activity. Last activity: .