KEKeyboard Community
Community homeHot discussions Latest posts Best of Knowledge base Resources About

Keyboard Community · Building an internal platform with keyboard from scratch: architecture and decision log

269
KEr/keyboard·posted by li_ming·2 days agoPostmortem

Building an internal platform with keyboard from scratch: architecture and decision log

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

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.

Mmysqlserverteam.comExternal link · opens in a new tab
55 comments

55 comments

M
Aalice_devOP·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.

482
Kkite·2 days agoedited

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

303
Ddev_zhou·yesterday

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.

12
Nnikic·3 minutes agoedited

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

502
Cchen_dev·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.

58
Bbob_chen·2 days ago

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

427
Aalice_dev·5 hours ago

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

371
Sslow_query·just nowedited

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.

156
Rrase·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.

44
Kkernel_panic·5 hours agoedited

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

257
Sswoole_leeOP·1 hour 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.

390
Bbob_chenMod·2 days ago

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

66
Aalice_dev·yesterday

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

1
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.

302
Aalice_dev·3 minutes agoedited

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

296
Hhuang_ke·1 hour ago

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

432
Lli_mingOP·12 minutes ago

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

292
Mmike_xu·2 days ago

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

491
Zzhu_zongMod·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.

350
Kkite·5 hours agoeditedLevel 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.

101
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.

27
Wwinter·2 days agoLevel 6

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

184
Ddev_zhou·3 minutes agoLevel 6

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.

154
Ddev_zhouMod·12 minutes agoedited

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

432
Aalice_devOP·2 days ago

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

260
Mmike_xu·28 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.

155
Ddev_zhou·12 minutes 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.

93
Rrase·2 days ago

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

132
Kkernel_panicOPMod·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.

21
Kkite·2 days agoLevel 6

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

115
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.

101
Zzhu_zong·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.

56
Oops_wang·2 hours ago

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

279
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.

386
Cchen_dev·2 days ago

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

76
Wwinter·2 days ago

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

247
Ttang_hao·2 days ago

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

221
Oops_wang·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.

218
Rran_boMod·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.

395
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.

126
Rrase·2 days ago

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

121
Llinlin·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.

112
Kkite·2 days ago

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

92
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.

85
Hhuang_keOP·just now

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

22
Aalice_dev·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.

32
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.

25
Wwinter·3 minutes 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.

20
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.

7
Sslow_query·just now

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

1
Rrase·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.

146
Lli_ming·yesterday

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

249
Hhuang_ke·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.

70
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.

40
Rrase·2 days ago

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

1

This is the post detail page /en/c/keyboard/post/p10. 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 →