pageserver produces page read errors (`ECANCELED`) (presumably due to SIGTERM)
Guide pageserver produces page read errors (`ECANCELED`) (presumably due to SIGTERM): [alexanderlaw (reporter)]: > But note my comment > > Tried 6.14.5 (latest LTS), which has this patch. Doesn't repro. > However, it also doesn't repro on 6.14.5 with this patch reverted. Use this when you hit exactly this in pageserver. Not for different errors or different tools.
TL;DR: [alexanderlaw (reporter)]: > But note my comment > > Tried 6.14.5 (latest LTS), which has this patch. Doesn't repro. > However, it also doesn't repro on 6.14.5 with this patch reverted. > Yes, it corresponds to my find: it looks like the issue was fixed in kernel version 6.13.8. UPD: Sorry, I hadn't looked at the link, thought you referred a commit made for 6.14.5.
Fix
[alexanderlaw (reporter)]: > But note my comment > > Tried 6.14.5 (latest LTS), which has this patch. Expected: this specific failure stops.
Doesn't repro. > However, it also doesn't repro on 6.14.5 with this patch reverted. > Yes, it corresponds to my find: it looks like the issue was fixed in kernel version 6.13.8. Expected: this specific failure stops.
UPD: Sorry, I hadn't looked at the link, thought you referred a commit made for 6.14.5. Expected: this specific failure stops.
Context: GitHub issue neondatabase/neon#11446 (closed, 21 comments): When reproducing "could not ingest record" (YOURHOST.com/neondatabase/neon/issues/11427) (I ran 15 instances of testpageserverrestartsunder_worload[release-pg17] in parallel), besides those errors, I also got another interesting one:. Expected: this specific failure stops.
Details
[alexanderlaw (reporter)]: > But note my comment > > Tried 6.14.5 (latest LTS), which has this patch. Doesn't repro. > However, it also doesn't repro on 6.14.5 with this patch reverted. > Yes, it corresponds to my find: it looks like the issue was fixed in kernel version 6.13.8. UPD: Sorry, I hadn't looked at the link, thought you referred a commit made for 6.14.5. Context: GitHub issue neondatabase/neon#11446 (closed, 21 comments): When reproducing "could not ingest record" (YOURHOST.com/neondatabase/neon/issues/11427) (I ran 15 instances of testpageserverrestartsunderworload[release-pg17] in parallel), besides those errors, I also got another interesting one: ` 9 2025-04-02 15:42:11.256 INFO [utils.py:514] not allowed pageserver1 error: 2025-04-02T12:41:43.352467Z ERROR pageserviceconnmain{peeraddr=the loopback address:50098 applicationname=2568761 computemode=primary}:processquery{tenantid=868bb64929f843d98ee2463e3749566c timelineid=576bfa03d129ba39d0d484b551825c3c}:handlepagerequests:request:handlegetpag
When to use
You hit exactly this: pageserver produces page read errors (ECANCELED) (presumably due to SIGTERM) in pageserver.
When not to use
A different error, or the same symptom in a different tool. This page only covers the failure above.
Compatibility
pageserver (versions mentioned here: 6.14.5, 6.13.8, 11.256).