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

Capacitor · Ops · Discussion: is the capacitor-ops ecosystem being replaced by another stack?

801
CAr/capacitor-ops·posted by tang_hao·yesterdayTutorial

Discussion: is the capacitor-ops ecosystem being replaced by another stack?

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

On trade-offs, my view is this: if nobody on the team owns this area long-term, do not introduce a second mechanism. With two coexistence you first have to work out which one is even in play when things break, and that costs far more than the performance you saved.

The first thing was to collapse the variables. We were changing config and upgrading the version at the same time, and afterwards nobody could say which change caused what. We rolled back to moving one variable at a time, re-ran three times, and only then did the curve settle. Tedious, but not skippable.

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.

302 comments

302 comments

· first 120 loaded
M
Zzhu_zong·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.

496
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.

481
Oops_wang·3 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.

476
Zzhu_zong·12 minutes ago

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

470
Ddev_zhou·12 minutes ago

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

10
Llinlin·2 days ago

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

455
Lli_mingOP·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.

306
Kkite·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.

150
Sslow_query·2 hours agoLevel 6

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.

388
Zzhu_zong·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.

67
Llinlin·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.

104
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.

172
Zzhu_zongOP·2 days ago

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

21
WwinterMod·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.

40
LlinlinOP·2 days ago

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

56
Ttang_hao·2 days ago

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

1
Wwinter·2 days ago

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

311
Lli_ming·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.

2
Zzhu_zong·1 hour 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.

454
Aalice_dev·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.

105
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.

93
WwinterMod·just nowedited

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.

368
Cchen_dev·2 days ago

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

368
Ddev_zhou·12 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.

359
Zzhou_yi·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.

323
Sslow_queryOP·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.

213
Mmike_xu·2 days ago

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

208
Zzhou_yi·2 days ago

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

273
Mmike_xu·2 days ago

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

249
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.

70
Llinlin·3 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.

248
Rrase·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.

198
Cchen_dev·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.

237
Zzhou_yi·2 days ago

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

175
Zzhou_yi·2 days ago

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

171
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.

171
Lli_ming·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.

157
Rran_bo·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.

392
Sslow_queryOPMod·2 days ago

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

41
Wwinter·2 days ago

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

79
Ddev_zhou·5 hours ago

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

155
Ttang_haoOP·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.

77
Oops_wang·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.

338
Ddev_zhou·2 days agoedited

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

199
Sswoole_lee·2 days ago

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

66
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.

136
Sslow_query·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.

381
Sslow_query·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.

3
Mmike_xu·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.

116
Mmike_xuOP·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.

63
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.

114
Llinlin·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.

1
Kkite·2 hours ago

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

108
Rrase·2 days agoedited

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

74
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.

84
Kkite·2 days ago

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

27
Bbob_chen·2 days ago

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

1
Ddev_zhou·2 days ago

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

92
Cchen_dev·12 minutes ago

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

87
RraseOP·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.

23
Nnikic·28 minutes ago

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

82
Nnikic·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.

80
Oops_wangOP·2 days ago

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

16
Lli_mingOP·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.

397
Ddev_zhouMod·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.

52
Llinlin·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.

78
Zzhu_zongMod·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.

73
Rran_boOP·12 minutes ago

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

12
Hhuang_ke·2 days ago

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

51
Rran_bo·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.

47
Wwinter·just now

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.

23
Wwinter·2 days ago

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

17
Rran_bo·3 minutes ago

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

5
Rran_bo·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.

208
Kkernel_panic·3 minutes agoedited

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

59
Nnikic·28 minutes agoLevel 6

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.

2
Aalice_dev·5 hours agoedited

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

5
Llinlin·2 days agoLevel 6

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

175
Ddev_zhou·2 days agoLevel 6

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

61
Rrase·2 days agoedited

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

119
Bbob_chenMod·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.

12
Ddev_zhou·yesterday

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

478
Mmike_xuOP·2 hours ago

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

172
Mmike_xuMod·2 days agoLevel 6

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

487
Sslow_queryOP·3 minutes ago

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

94
Rran_bo·3 minutes ago

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

36
Sslow_query·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.

301
Hhuang_keOP·2 days agoLevel 6

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

8
Rrase·2 days agoLevel 6

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

56
Lli_ming·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.

17
Rran_bo·12 minutes agoeditedLevel 6

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

1
Rrase·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.

18
Sswoole_leeMod·2 hours agoLevel 6

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.

2
Zzhu_zong·5 hours agoeditedLevel 6

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

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.

3
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.

46
Kkite·12 minutes ago

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

41
Zzhu_zong·2 days ago

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

188
Nnikic·2 days ago

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

16
Cchen_dev·2 days ago

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

1
Llinlin·2 days agoedited

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

90
Wwinter·2 days ago

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

32
Ttang_hao·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.

17
Rran_bo·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.

17
Kkernel_panicMod·2 days ago

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

135
Kkernel_panic·1 hour 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.

24
Nnikic·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.

126
WwinterMod·2 days ago

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

58
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.

16
Oops_wang·3 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.

7
Rrase·2 days agoedited

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

6
Sswoole_lee·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.

1
Ddev_zhou·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.

380
Lli_ming·2 days ago

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

1
Lli_ming·2 days ago

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

338
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.

13
Kkernel_panic·2 days ago

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

1
Lli_ming·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.

193
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.

206
Hhuang_ke·2 days ago

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

1

This is the post detail page /en/c/capacitor-ops/post/p2. 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 →