TUTurborepo · Job board
Community homeHot discussions Latest posts Best of Knowledge base Resources About

Turborepo · Job board · Built a visual debugging panel for turborepo-jobs over the weekend

127
TUr/turborepo-jobs·posted by nikic·32 minutes agoDiscussion

Built a visual debugging panel for turborepo-jobs over the weekend

Short version: turborepo-jobs 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.

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.

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

Zzhihu.comExternal link · opens in a new tab
60 comments

60 comments

M
LlinlinOP·5 hours agoedited

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

432
Rrase·2 days agoedited

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

481
Rrase·28 minutes ago

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

429
Zzhou_yi·28 minutes ago

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

9
Oops_wang·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
NnikicOP·12 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.

1
Sslow_query·2 hours ago

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

1
Kkernel_panic·2 days agoedited

One counter-example: below turborepo-jobs 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
Zzhu_zong·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.

388
Sswoole_lee·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.

37
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.

265
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.

121
Sswoole_lee·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.

304
Bbob_chen·1 hour ago

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

369
Sswoole_lee·2 days ago

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

93
Hhuang_ke·2 days ago

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

350
Lli_ming·1 hour 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.

250
Ttang_hao·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.

248
Bbob_chen·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.

150
Zzhu_zong·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.

309
Oops_wang·yesterday

One counter-example: below turborepo-jobs 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
Ttang_hao·28 minutes ago

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

357
Hhuang_keMod·2 days ago

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

37
Bbob_chen·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.

143
Cchen_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.

308
Ddev_zhou·12 minutes agoedited

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

7
Hhuang_keMod·2 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.

1
Zzhou_yi·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.

2
RraseMod·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.

1
Oops_wang·1 hour ago

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

196
Wwinter·2 days ago

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

488
Bbob_chen·2 hours agoLevel 6

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

137
Zzhou_yiMod·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.

57
Kkernel_panic·2 days agoLevel 6

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

15
Sswoole_lee·12 minutes ago

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

23
Hhuang_ke·2 days agoedited

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

19
Ttang_hao·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.

253
Sswoole_lee·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.

329
Kkernel_panic·2 days ago

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

139
Ttang_hao·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.

124
Sslow_query·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.

118
Mmike_xu·yesterday

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

114
Nnikic·2 days ago

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

198
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.

85
Ttang_hao·12 minutes ago

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

107
Sswoole_lee·3 minutes ago

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

99
Ttang_haoMod·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.

67
Lli_ming·yesterday

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.

66
Aalice_devOP·1 hour ago

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

1
RraseOP·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.

3
KkiteMod·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.

21
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.

20
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.

7
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.

310
Zzhou_yiMod·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.

399
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.

337
Zzhou_yi·5 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.

198
Ddev_zhou·yesterday

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.

32
Oops_wang·2 days ago

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

2
Ddev_zhou·2 days agoedited

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

5

This is the post detail page /en/c/turborepo-jobs/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 →