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

Alertmanager · English · Help: alertmanager-en fails to start intermittently in containers, logs inside

617
ALr/alertmanager-en·posted by tang_hao·32 minutes agoTutorial

Help: alertmanager-en fails to start intermittently in containers, logs inside

Short version: alertmanager-en 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.

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".

# Load test: do not jump straight to the max concurrency.
# Ramp it up, otherwise you miss the knee.
for c in 50 100 200 400 800; do
  wrk -t8 -c$c -d60s --latency http://127.0.0.1:8080/api/feed
  sleep 20
done

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.

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.

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.

315 comments

315 comments

· first 120 loaded
M
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.

481
Zzhu_zongOP·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.

246
Nnikic·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.

97
Cchen_dev·5 hours ago

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

461
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.

24
Rran_bo·just now

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

451
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.

359
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.

451
Kkite·1 hour 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.

446
Wwinter·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.

413
Kkernel_panic·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.

224
Cchen_devOP·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.

333
Kkite·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.

260
Hhuang_ke·1 hour 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.

142
Bbob_chen·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.

328
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.

391
Zzhu_zong·2 days ago

One counter-example: below alertmanager-en 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
Sswoole_leeOP·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.

1
Mmike_xu·2 days ago

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

381
Kkite·2 days ago

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

373
Zzhu_zong·2 days ago

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

315
Rran_boOP·28 minutes ago

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

239
Nnikic·2 days agoedited

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

160
Aalice_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.

480
Cchen_dev·2 days ago

I just read the alertmanager-en 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

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

85
Oops_wang·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.

16
Rran_bo·2 days agoeditedLevel 6

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

456
Mmike_xu·2 days 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.

9
Cchen_dev·yesterdayedited

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

17
NnikicOPMod·3 minutes ago

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

202
Mmike_xu·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.

259
Rran_bo·2 days ago

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

245
Hhuang_ke·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.

233
Ttang_hao·2 days ago

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

197
Llinlin·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.

191
Llinlin·12 minutes ago

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

187
Ttang_hao·just now

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

179
Cchen_dev·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.

252
Hhuang_ke·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.

39
Ttang_hao·2 days ago

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

136
Mmike_xu·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.

97
Sswoole_lee·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.

49
Llinlin·3 minutes ago

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

1
Sslow_query·2 days ago

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

93
Rran_boMod·2 days ago

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

175
Kkernel_panic·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.

139
Rrase·28 minutes agoedited

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

438
Kkite·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.

456
Mmike_xu·2 days ago

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

2
Cchen_dev·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.

1
Llinlin·2 days agoeditedLevel 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.

9
Mmike_xu·2 days 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
Mmike_xu·2 days ago

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

357
Nnikic·yesterday

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.

190
Kkite·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.

399
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.

85
Aalice_dev·2 days ago

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

453
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.

10
Ttang_hao·2 days agoedited

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

18
Sswoole_lee·2 days ago

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

64
Sslow_queryMod·2 days agoedited

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

173
Kkite·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.

138
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.

279
Wwinter·2 days ago

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

340
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.

279
Ddev_zhou·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.

211
Kkernel_panic·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.

5
Bbob_chen·3 minutes agoedited

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

9
Zzhu_zong·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.

73
Nnikic·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.

2
Nnikic·2 days ago

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

214
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.

115
Rran_bo·2 days ago

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

124
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.

111
Sswoole_lee·2 days ago

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

107
NnikicMod·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.

107
Oops_wang·2 days ago

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

97
Llinlin·2 days ago

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

92
Zzhou_yi·2 days ago

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

373
LlinlinOP·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.

30
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.

85
Sslow_query·3 minutes ago

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

161
Zzhu_zong·1 hour ago

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

79
Sslow_query·2 days ago

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

167
Rrase·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.

115
Kkite·2 days ago

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

30
Cchen_dev·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.

76
Sswoole_lee·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.

69
Wwinter·2 days ago

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

64
Rrase·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.

128
Rran_boOPMod·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
Mmike_xu·2 days ago

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

15
Lli_ming·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.

36
RraseOP·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.

4
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.

400
Zzhou_yi·2 days ago

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

141
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.

71
Sswoole_lee·2 hours agoedited

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

31
Kkite·2 days ago

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

26
Kkernel_panic·2 days ago

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

24
Lli_ming·2 days ago

One counter-example: below alertmanager-en 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_lee·2 days ago

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

21
Wwinter·2 days ago

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

13
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.

160
Sslow_queryMod·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.

4
Rran_bo·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.

5
Kkernel_panic·2 days ago

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

11
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.

10
Sswoole_lee·12 minutes 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.

7
Oops_wang·2 days ago

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

8
Lli_mingMod·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.

317
Rran_bo·2 days agoedited

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

6
Zzhou_yi·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.

6
Sswoole_leeMod·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.

3
Zzhou_yi·yesterday

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
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.

247
Nnikic·2 days ago

This is not a alertmanager-en 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

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

1
Hhuang_ke·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.

1

This is the post detail page /en/c/alertmanager-en/post/p13. 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 →