IOiOS · Open source
Community homeHot discussions Latest posts Best of Knowledge base Resources About

iOS · Open source · This ios-opensource bug hid for three years before anyone reported it

735
IOr/ios-opensource·posted by chen_dev·6 hours agoJobs

This ios-opensource bug hid for three years before anyone reported it

Short version: ios-opensource needs almost no tuning at small and medium scale — the point where it starts to hurt is much further out than most people assume. Full measurements below.

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.

One last trap: in container environments remember to adjust the memory-related parameters in step. Otherwise the host limit and the process expectation disagree, and the symptom is intermittent, unreproducible failure.

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.

169 comments

169 comments

· first 120 loaded
M
Aalice_dev·2 days agoedited

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

509
Kkernel_panic·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.

494
Bbob_chen·2 days ago

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

17
Llinlin·2 days ago

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

131
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.

481
Sslow_query·2 days agoedited

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

451
Nnikic·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.

428
Zzhu_zong·2 days ago

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

419
Wwinter·2 days ago

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

406
Kkernel_panic·2 days ago

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

402
Ttang_hao·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.

370
Mmike_xu·2 days ago

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

370
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.

157
Cchen_dev·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.

362
Bbob_chen·2 days ago

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

12
WwinterOP·2 days ago

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

297
Sslow_query·3 minutes ago

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

295
Rrase·2 days ago

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

294
Bbob_chen·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.

261
Llinlin·12 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.

251
Sswoole_lee·2 days ago

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

247
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.

247
Hhuang_ke·2 days ago

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

238
Cchen_dev·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.

217
LlinlinOP·2 days ago

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

131
Aalice_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.

170
Kkernel_panic·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.

170
Zzhu_zong·2 days ago

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

161
Sswoole_lee·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.

128
Lli_mingOP·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.

60
Kkernel_panic·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.

213
Lli_ming·12 minutes 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.

303
Lli_ming·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.

515
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.

130
Sswoole_lee·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.

218
Sslow_query·28 minutes ago

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

1
Sslow_query·2 days agoLevel 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.

465
Cchen_dev·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.

10
Aalice_dev·1 hour agoeditedLevel 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.

5
Hhuang_ke·2 days ago

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

1
Rrase·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.

490
Bbob_chen·2 days ago

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

5
Oops_wang·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.

219
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.

123
Sslow_queryMod·2 days ago

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

354
Kkite·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.

118
Bbob_chen·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.

255
Rran_bo·2 days ago

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

112
Rran_boMod·2 days ago

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

414
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.

269
Zzhou_yi·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.

66
Oops_wangMod·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.

229
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.

6
Nnikic·12 minutes agoedited

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

82
Ddev_zhouOP·2 days ago

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

17
Wwinter·2 days ago

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

219
Ttang_hao·5 hours ago

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

73
Rrase·2 days ago

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

65
LlinlinOP·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.

1
Bbob_chen·2 days ago

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

60
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.

57
Wwinter·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.

3
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.

48
Cchen_dev·2 days ago

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

98
Aalice_dev·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.

173
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.

61
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.

42
Oops_wang·2 days ago

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

41
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.

3
Sslow_query·2 days ago

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

41
Hhuang_ke·2 hours agoedited

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

39
Aalice_dev·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.

30
Mmike_xu·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.

28
Llinlin·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.

426
Mmike_xu·2 days ago

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

6
Hhuang_ke·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
Cchen_dev·2 days ago

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

26
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.

407
Aalice_dev·2 days ago

This is not a ios-opensource 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

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.

414
Mmike_xu·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.

367
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.

100
Kkite·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.

20
Rrase·2 days ago

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

21
Rran_bo·2 days ago

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

194
Hhuang_ke·2 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.

15
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.

15
Nnikic·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.

14
Llinlin·2 days ago

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

484
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.

94
Aalice_devMod·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.

41
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.

14
Rrase·1 hour ago

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

13
Sslow_query·3 minutes ago

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

11
Mmike_xu·12 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.

31
Lli_ming·2 days ago

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

9
Aalice_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.

5
Oops_wang·1 hour ago

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

4
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.

254
Ttang_hao·2 days ago

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

228
Sslow_queryOP·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.

57
Nnikic·5 hours agoedited

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

282
Zzhu_zong·2 days agoedited

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

262
Cchen_dev·3 minutes ago

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

43
Lli_ming·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.

514
Wwinter·3 minutes agoLevel 6

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

79
Sslow_query·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.

63
Zzhu_zong·2 days agoLevel 6

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

416
Bbob_chen·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.

43
Sslow_query·2 days ago

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

2
Rrase·2 days ago

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

2
Lli_ming·2 days agoLevel 6

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

1
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.

89
Hhuang_ke·3 minutes ago

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

3
Aalice_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.

2
Rran_bo·yesterday

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

1
Rrase·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.

21
Nnikic·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.

1
Bbob_chen·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.

1
Hhuang_keMod·2 days agoedited

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

1

This is the post detail page /en/c/ios-opensource/post/p11. 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 →