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

SWC · Interviews · Postmortem: how we exhausted the connection pool in swc-interview

1.5K
SWr/swc-interview·posted by kernel_panic·3 hours agoReview

Postmortem: how we exhausted the connection pool in swc-interview

Most swc-interview articles stop at "how to use it" and never cover "when not to use it". This is an attempt at the second half.

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 swc-interview itself but our upstream connection reuse — the load test traffic was too clean and hid the long-tail requests.

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

578 comments

· first 120 loaded
M
Kkernel_panic·3 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.

520
Lli_ming·2 hours ago

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

377
Ttang_hao·28 minutes agoedited

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

212
Mmike_xu·2 days agoedited

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

154
Hhuang_ke·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.

106
Hhuang_ke·2 days ago

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

163
Mmike_xu·2 days agoLevel 6

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

201
Zzhu_zong·2 days ago

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

1
Ttang_hao·28 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.

55
Ddev_zhou·12 minutes 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.

87
Kkite·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.

1
Wwinter·2 days ago

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

68
Rran_bo·2 days agoLevel 6

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.

488
Lli_ming·2 days agoLevel 6

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

472
Lli_ming·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.

37
Bbob_chen·2 days agoLevel 6

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

1
Ddev_zhou·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.

39
Rran_bo·2 days ago

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

114
Ttang_hao·2 days agoedited

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

72
Mmike_xu·2 days ago

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

1
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.

162
Hhuang_ke·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.

33
Ttang_haoMod·2 hours 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.

490
Kkernel_panic·5 hours 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.

475
Hhuang_ke·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.

475
Bbob_chen·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.

442
Wwinter·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.

418
Wwinter·just now

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

368
Hhuang_ke·2 days ago

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

454
Llinlin·2 days ago

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

62
Cchen_dev·12 minutes ago

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

3
Hhuang_ke·28 minutes ago

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

363
Sswoole_lee·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.

322
Nnikic·2 days ago

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

79
Wwinter·12 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.

295
Bbob_chen·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.

390
Oops_wangMod·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.

344
Nnikic·2 days ago

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

318
Ddev_zhou·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.

65
Rrase·2 days agoedited

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.

119
Sslow_query·2 days ago

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

1
Cchen_dev·2 days ago

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

315
Bbob_chen·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.

264
Cchen_dev·12 minutes ago

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

219
Cchen_dev·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.

16
Llinlin·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.

249
Aalice_dev·2 days agoedited

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.

232
Lli_mingOP·2 days ago

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

169
Bbob_chenOP·2 days ago

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

164
Rran_bo·2 days agoedited

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.

357
Ddev_zhouOP·28 minutes 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.

161
Mmike_xu·2 days agoedited

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

502
Cchen_dev·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.

161
Lli_ming·28 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.

141
Rran_bo·2 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.

88
Ttang_hao·2 days ago

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

37
NnikicOPMod·2 days ago

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

155
Rrase·2 days agoedited

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.

2
Zzhu_zong·3 minutes ago

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

197
RraseOP·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.

126
Zzhu_zong·2 days agoedited

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

181
Aalice_dev·2 days ago

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

1
Sslow_query·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.

170
Zzhou_yi·2 days ago

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

140
Rrase·5 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.

140
Cchen_devMod·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.

400
Lli_ming·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.

128
Ddev_zhou·2 days ago

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

1
Zzhou_yiOP·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.

59
Bbob_chen·2 days agoedited

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

110
Sswoole_lee·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.

69
Zzhu_zong·2 days ago

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

69
Ddev_zhouOP·2 days agoedited

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.

8
Wwinter·5 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.

62
Kkite·2 days ago

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

61
Wwinter·12 minutes ago

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

449
Kkite·2 hours ago

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

275
Sslow_query·2 days ago

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

9
Bbob_chen·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.

61
Lli_ming·5 hours 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.

57
Ttang_hao·2 days ago

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

37
Sslow_query·12 minutes ago

One counter-example: below swc-interview 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
Cchen_dev·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.

31
Lli_ming·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.

27
Sswoole_lee·2 days ago

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

211
Kkite·just now

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

23
Sswoole_leeOP·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.

42
Sswoole_leeOP·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.

99
Mmike_xu·just now

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

19
Kkernel_panic·12 minutes ago

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

77
Nnikic·2 hours ago

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

501
Kkernel_panic·1 hour ago

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

484
Kkite·2 days ago

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

133
Llinlin·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.

58
Llinlin·2 days agoeditedLevel 6

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.

335
Mmike_xu·2 days agoLevel 6

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

243
Wwinter·2 days agoeditedLevel 6

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

164
Kkite·2 days ago

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

41
Sslow_query·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.

116
Rran_bo·2 days ago

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

78
Sslow_queryOP·2 days agoLevel 6

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.

486
Bbob_chen·2 days ago

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

42
Oops_wang·2 days agoLevel 6

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.

204
Rran_bo·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.

29
Aalice_dev·2 days agoLevel 6

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

3
Bbob_chen·12 minutes 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.

21
Wwinter·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.

450
Cchen_dev·2 days agoLevel 6

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

271
Kkernel_panic·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.

9
Oops_wang·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.

9
Ddev_zhou·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.

9
Hhuang_ke·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.

55
Aalice_dev·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.

6
Ddev_zhou·2 days ago

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

3
Ddev_zhou·yesterday

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

2
Hhuang_ke·2 hours ago

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

1
Cchen_dev·2 days ago

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

49
Nnikic·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.

1
Nnikic·3 minutes ago

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

1
Rrase·2 days agoedited

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

1

This is the post detail page /en/c/swc-interview/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 →