JUJupyter · Advanced
Community homeHot discussions Latest posts Best of Knowledge base Resources About

Jupyter · Advanced · Postmortem: how we exhausted the connection pool in jupyter-pro

96
JUr/jupyter-pro·posted by dev_zhou·2 hours agoTooling

Postmortem: how we exhausted the connection pool in jupyter-pro

It took me two weeks of on-and-off digging and plenty of wrong turns. Writing the process down as it happened so the next person spends less time.

We also fixed monitoring along the way: replaced average-based alerts with percentiles and split them per endpoint. False alerts dropped by about seventy percent and the on-call rotation visibly cheered up.

Image placeholder · object storage in production
POST /api/uploads → CDN origin pull
11 comments

11 comments

M
Zzhu_zong·1 hour ago

I just read the jupyter-pro source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

362
Kkernel_panic·5 hours ago

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

300
Lli_ming·1 hour ago

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

265
Nnikic·3 minutes ago

We have run this in production for two years without hitting it. That said, we never reached this scale, so our experience is not really evidence here.

215
Oops_wang·28 minutes agoedited

Worth learning from this debugging approach. We went straight at the logs and took a much longer route.

118
Oops_wang·28 minutes ago

Has anyone run a controlled experiment? I did, reducing it to a single variable, and the difference was 4% — within noise. So I suspect the main cause is something else.

48
Wwinter·yesterdayedited

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

264
Sslow_queryMod·5 hours ago

There is actually a simpler fix that needs no architecture change: move this check up to the gateway and the problem disappears. The cost is one extra lookup at the gateway.

57
Llinlin·2 hours agoedited

Agreeing with the above. One addition: with this option enabled the GC count in your metrics doubles, so adjust the alert threshold at the same time or it will keep firing.

149
Ttang_hao·3 minutes ago

This is not a jupyter-pro problem, it is a usage problem. The docs say this API is not thread-safe and you must lock around it yourself.

33
Ttang_hao·28 minutes ago

One counter-example: below jupyter-pro 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

4

This is the post detail page /en/c/jupyter-pro/post/p0. Posts and comments are generated deterministically from a seeded PRNG, so the same post always renders the same content and the link can be shared, reloaded and indexed. In production this page reads MySQL for the post, Redis for hot-post caching, and fetches the whole comment tree in a single query on the path column.

See the database schema →