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

Fortran · Resources · Why this fortran-resources pattern gets slow in production: a source-level explanation

141
FOr/fortran-resources·posted by bob_chen·1 hour agoInternals

Why this fortran-resources pattern gets slow in production: a source-level explanation

Some background first. Our setup is fortran-resources plus three downstream services, seven figures of daily requests, peaking around nine in the evening.

Order of investigation, by return on effort: 1. Check downstream latency first — usually it is not your problem 2. Then pool hit rate and wait-queue length 3. Only then GC and allocation 4. Suspect the framework last

What genuinely surprised me was the tail. The average looked great while P99 jumped by an order of magnitude past some threshold. The cause was not fortran-resources itself but our upstream connection reuse — the load test traffic was too clean and hid the long-tail requests.

Worth noting: the official docs do cover this, just in a very inconspicuous spot. I only found it reading the source comments, where the author explains the reasoning — roughly "so that it degrades into predictable behaviour in extreme cases".

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.

84 comments

84 comments

M
Ddev_zhou·2 days agoedited

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.

468
Sswoole_lee·2 days ago

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

228
Aalice_dev·2 days ago

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

466
Oops_wang·2 days 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.

415
Sslow_query·2 days ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

305
Lli_ming·just now

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

303
Oops_wang·2 days agoedited

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

22
Ddev_zhouOP·12 minutes ago

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

19
Wwinter·2 days ago

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

28
Kkite·2 days 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.

404
Nnikic·just now

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.

385
Rran_bo·2 days ago

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.

366
Llinlin·2 days ago

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

19
Ddev_zhou·2 days 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.

360
NnikicOP·just now

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

279
Aalice_devOPMod·2 days ago

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

36
Zzhu_zong·12 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.

211
Zzhu_zong·2 days 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.

2
Rrase·28 minutes ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

231
Sswoole_lee·2 days ago

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

22
Ttang_hao·2 days agoLevel 6

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

271
Zzhu_zong·2 days agoLevel 6

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.

10
Sswoole_lee·2 hours 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.

337
Rran_boOP·2 days ago

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

466
Zzhou_yi·2 days 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.

11
Rrase·2 days 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.

81
Aalice_dev·2 hours ago

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

3
Kkernel_panic·2 days ago

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

376
Wwinter·2 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.

117
Bbob_chen·2 days ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

163
Llinlin·2 days agoedited

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

5
Nnikic·just now

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

12
Cchen_devMod·2 days ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

155
Kkernel_panic·2 days ago

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

57
Rrase·2 days ago

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

284
Aalice_devMod·2 days ago

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

274
Zzhu_zong·2 days ago

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

248
Rran_bo·2 days ago

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

271
Ddev_zhou·2 days 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.

231
Wwinter·12 minutes ago

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.

209
Aalice_devOP·28 minutes ago

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

148
Wwinter·just now

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

55
Lli_ming·just now

Saved. I am reworking this area this week — this saves a lot of wrong turns.

2
Oops_wang·2 days agoedited

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

1
Nnikic·2 days ago

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

197
Rran_bo·1 hour ago

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

190
Ttang_hao·2 days ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

178
KkiteOP·2 days ago

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

96
Sswoole_lee·2 days ago

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.

1
Wwinter·2 days ago

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

129
Wwinter·2 days ago

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

233
Lli_ming·3 minutes agoedited

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

107
Bbob_chen·2 days ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

105
Nnikic·28 minutes 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.

93
Rran_bo·2 days 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.

348
Kkernel_panicOP·2 days ago

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

21
Cchen_dev·2 hours ago

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.

260
Sslow_query·2 days agoedited

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

81
Zzhou_yi·2 days agoedited

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.

25
Ddev_zhouOPMod·2 days ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

21
Zzhu_zong·2 days 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.

76
Hhuang_keOP·2 days ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

14
Ttang_hao·2 days ago

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

71
Rran_bo·5 hours ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

64
LlinlinMod·2 days ago

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.

57
Rrase·2 days 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.

430
Rran_boMod·2 days ago

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

41
Sswoole_lee·2 days ago

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

19
Zzhou_yi·2 days 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.

11
Sswoole_lee·2 days ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

439
Lli_ming·yesterday

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.

4
Zzhu_zong·2 days 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.

11
Bbob_chen·12 minutes ago

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

7
Rrase·5 hours agoedited

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.

13
Ttang_haoOP·just now

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.

17
Rrase·just now

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

5
Kkite·2 days ago

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

7
Hhuang_keOP·1 hour ago

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

444
Rrase·2 days ago

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

71
Wwinter·2 days ago

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

219
Oops_wang·2 days ago

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

33
Rran_bo·12 minutes ago

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

3
Sslow_query·2 days 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.

1
Sslow_query·1 hour agoedited

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

1

This is the post detail page /en/c/fortran-resources/post/p5. 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 →