POPolars · Beginners
Community homeHot discussions Latest posts Best of Knowledge base Resources About

Polars · Beginners · I built a small tool with polars-newbie — code is open, feedback welcome

961
POr/polars-newbie·posted by winter·1 hour agoAnnouncement

I built a small tool with polars-newbie — code is open, feedback welcome

It took me two weeks of on-and-off digging and plenty of wrong turns. Writing the process down as it happened so the next person spends less time.

# 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.

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.

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.

525 comments

525 comments

· first 120 loaded
M
Zzhou_yi·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.

506
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.

462
Ttang_hao·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.

3
Zzhu_zong·2 days ago

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

119
Kkernel_panic·2 days ago

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

58
Zzhu_zongOP·2 days ago

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

372
Zzhou_yi·2 days ago

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

404
Lli_ming·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.

355
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.

336
Mmike_xu·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.

236
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.

320
Nnikic·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.

42
Lli_ming·3 minutes ago

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

316
Ddev_zhouMod·2 days ago

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

296
Kkernel_panic·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.

287
Sswoole_lee·2 days ago

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

269
Rrase·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.

227
Zzhu_zong·2 days ago

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

217
Rrase·28 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.

200
WwinterOP·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.

78
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.

424
Llinlin·2 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.

198
Sswoole_lee·just now

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

136
Bbob_chen·28 minutes agoedited

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

129
Zzhu_zong·2 days agoedited

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

128
Rrase·2 days agoedited

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

121
Sswoole_lee·just now

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.

116
Wwinter·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.

86
Mmike_xu·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.

259
Zzhou_yi·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.

83
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.

78
Mmike_xuOP·3 minutes ago

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

9
Llinlin·2 days ago

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

234
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.

66
WwinterOP·2 days ago

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

6
Sswoole_leeOP·2 days ago

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

2
Nnikic·5 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.

61
Sswoole_leeMod·2 days ago

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

27
Wwinter·2 days ago

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

21
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.

474
Llinlin·2 days ago

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

193
Wwinter·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
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.

28
Llinlin·2 hours ago

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

23
Sslow_query·3 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.

299
Kkernel_panic·12 minutes agoLevel 6

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

305
Ddev_zhou·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.

22
Cchen_dev·12 minutes ago

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

410
KkiteMod·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.

27
Ddev_zhou·5 hours 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
Kkernel_panic·2 days ago

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

252
Llinlin·2 hours ago

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

196
Oops_wang·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.

54
Bbob_chen·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.

288
Llinlin·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.

24
Ddev_zhou·2 hours 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.

19
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.

106
Sslow_query·12 minutes ago

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

440
Sswoole_lee·2 days ago

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

18
Sslow_queryMod·2 days ago

One counter-example: below polars-newbie 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
Llinlin·2 hours agoedited

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

15
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.

30
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.

3
Bbob_chen·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.

308
Sslow_query·2 days agoedited

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

305
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.

30
Oops_wangOP·2 hours ago

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

414
Wwinter·2 days ago

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

15
RraseMod·28 minutes agoedited

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

2
Sswoole_lee·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.

13
Cchen_dev·12 minutes ago

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

1
Oops_wangMod·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.

474
Lli_mingMod·28 minutes ago

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

96
LlinlinOP·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.

23
Bbob_chen·2 hours 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.

57
Hhuang_ke·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.

346
Zzhou_yi·2 days agoLevel 6

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

54
Sslow_query·2 days ago

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

29
Nnikic·2 days agoLevel 6

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

371
Wwinter·yesterday

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

24
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.

18
Oops_wangOP·2 days ago

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

440
Cchen_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.

78
Sswoole_lee·3 minutes agoedited

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

36
Wwinter·2 days agoLevel 6

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.

44
Zzhou_yi·2 hours agoedited

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

6
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.

317
Ttang_hao·yesterday

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

345
Zzhou_yi·2 days ago

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

33
Lli_ming·2 days ago

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

112
Hhuang_ke·2 hours 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.

24
Ttang_hao·2 days agoLevel 6

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
Aalice_dev·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.

307
Cchen_dev·2 days ago

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

424
Oops_wangOP·2 days ago

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

178
Sslow_query·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.

404
Oops_wangMod·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.

11
Zzhu_zong·2 days ago

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

237
Ddev_zhou·2 days ago

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

289
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.

217
Aalice_devOP·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.

154
Zzhu_zong·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.

1
Rrase·just nowLevel 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.

31
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.

184
Sswoole_lee·2 days ago

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

103
Kkite·2 days agoedited

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

6
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.

91
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.

80
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.

328
Sswoole_lee·2 days ago

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

417
Lli_ming·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.

262
Lli_ming·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.

24
Mmike_xu·1 hour ago

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

17
Sslow_query·2 days ago

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

10
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.

464
Mmike_xu·3 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.

1
Aalice_devMod·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.

278
Kkernel_panicMod·2 days ago

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

74
Mmike_xu·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
Aalice_dev·2 days ago

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

1

This is the post detail page /en/c/polars-newbie/post/p1. 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 →