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

Tanstack · Ops · Why this tanstack-ops pattern gets slow in production: a source-level explanation

415
TAr/tanstack-ops·posted by rase·1 hour agoTutorial

Why this tanstack-ops pattern gets slow in production: a source-level explanation

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

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.

-- The query that broke: a full scan over 20M rows.
-- A composite index took P99 from 1.8s down to 42ms.
SELECT id, title, created_at
  FROM posts
 WHERE community_id = ?
   AND status = 1
 ORDER BY score DESC
 LIMIT 20;

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.

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.

132 comments

132 comments

· first 120 loaded
M
Zzhu_zongOP·12 minutes ago

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

518
Ttang_hao·2 days ago

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

513
Sslow_query·2 days ago

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

508
Sswoole_lee·3 minutes ago

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

486
Rran_bo·3 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.

335
Rran_bo·2 days ago

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

27
Rran_bo·2 days ago

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

462
Kkernel_panic·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.

403
Ddev_zhou·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.

401
Oops_wang·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.

388
Ttang_hao·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.

367
Rran_bo·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.

365
Zzhu_zong·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.

358
Zzhou_yi·2 days ago

One counter-example: below tanstack-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.

352
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.

328
LlinlinOP·12 minutes ago

One counter-example: below tanstack-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.

242
Kkite·just now

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.

247
Aalice_dev·3 minutes ago

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

310
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.

2
Bbob_chenOPMod·2 days ago

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

99
Bbob_chen·yesterday

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.

27
Kkite·2 days ago

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

24
Bbob_chenMod·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.

119
Lli_ming·2 days ago

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

183
Rrase·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.

407
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.

293
Zzhou_yi·2 days agoLevel 6

One counter-example: below tanstack-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.

1
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.

188
Zzhu_zong·yesterdayedited

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.

108
Zzhu_zong·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.

1
Bbob_chenMod·2 days agoedited

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

32
Ddev_zhouOP·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.

233
Ttang_haoOP·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.

197
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.

255
Nnikic·1 hour ago

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

246
Ttang_hao·2 days ago

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

93
Mmike_xuMod·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.

174
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.

185
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.

23
Sslow_query·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.

240
Oops_wang·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.

229
Aalice_devMod·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.

205
Wwinter·2 days ago

One counter-example: below tanstack-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.

197
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.

193
Rrase·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.

185
Kkernel_panic·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.

519
Hhuang_ke·2 days ago

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

351
Rran_bo·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.

34
Wwinter·2 days ago

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

2
Rran_bo·2 days agoedited

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

104
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.

244
Lli_ming·2 days ago

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

8
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.

406
Aalice_dev·2 days ago

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

9
Zzhou_yiMod·2 days ago

One counter-example: below tanstack-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.

53
Rrase·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.

180
Cchen_devOP·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.

107
Nnikic·2 days ago

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

151
Zzhou_yi·2 days ago

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

149
Rrase·1 hour 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.

114
Kkernel_panic·2 days ago

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

108
Hhuang_ke·2 days ago

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

97
Wwinter·2 days ago

One counter-example: below tanstack-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.

92
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.

518
Zzhou_yi·28 minutes ago

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

150
Cchen_dev·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.

92
Sslow_query·12 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.

91
Mmike_xu·1 hour ago

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

159
Aalice_dev·2 hours 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.

418
Oops_wangOP·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.

32
Wwinter·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.

462
Rrase·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.

435
WwinterOP·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.

358
Wwinter·2 days ago

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

133
Llinlin·5 hours 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.

2
Zzhu_zong·2 days ago

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

87
Hhuang_ke·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.

507
Rran_bo·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.

120
RraseOP·2 days agoLevel 6

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

30
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.

2
Nnikic·2 days agoedited

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

5
Wwinter·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.

94
Sslow_query·yesterday

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

80
Ttang_hao·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.

302
Lli_mingOPMod·2 hours ago

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

17
Mmike_xu·2 days agoedited

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

62
Aalice_dev·2 days ago

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

55
Sswoole_lee·2 days ago

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

38
Cchen_dev·2 days ago

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

31
Aalice_devMod·2 days ago

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

360
Sswoole_lee·2 hours ago

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

28
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.

1
Oops_wang·2 days ago

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

368
Sslow_query·2 hours ago

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

380
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.

21
Wwinter·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.

65
Rran_bo·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.

242
Sslow_query·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.

214
Zzhou_yi·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.

208
Bbob_chenOP·5 hours ago

One counter-example: below tanstack-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.

12
Bbob_chen·2 days agoedited

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

390
Rran_bo·2 days 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.

14
Kkite·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.

58
Aalice_devOP·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.

19
Lli_ming·2 days agoedited

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

292
Ddev_zhou·2 days ago

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

117
Zzhu_zong·3 minutes ago

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

57
Ddev_zhouMod·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.

1
Mmike_xu·5 hours agoedited

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

15
Bbob_chenMod·28 minutes ago

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

8
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.

2
Sswoole_lee·5 hours ago

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

1
Kkite·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.

9
Ddev_zhou·2 days agoedited

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

64
Llinlin·2 days ago

One counter-example: below tanstack-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.

1
RraseMod·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.

399
Llinlin·2 days ago

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

28
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.

1
Oops_wang·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
Cchen_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.

1

This is the post detail page /en/c/tanstack-ops/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 →