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

Polars · Beginners · Building an internal platform with polars-newbie from scratch: architecture and decision log

667
POr/polars-newbie·posted by tang_hao·2 days agoTranslation

Building an internal platform with polars-newbie from scratch: architecture and decision log

Some background first. Our setup is polars-newbie 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.

Ddeveloper.mozilla.orgExternal link · opens in a new tab
383 comments

383 comments

· first 120 loaded
M
Bbob_chenOP·2 hours ago

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

487
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.

476
Lli_ming·2 days ago

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

399
RraseOPMod·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.

407
Rrase·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.

392
Sslow_query·1 hour agoedited

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

333
Zzhou_yi·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.

107
Ddev_zhou·2 days ago

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

133
Kkite·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.

329
Sslow_query·12 minutes ago

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

173
Lli_ming·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.

83
Sslow_query·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
Lli_ming·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.

324
Lli_ming·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.

1
Rran_bo·2 hours 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".

80
Nnikic·5 hours ago

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

275
Mmike_xu·just now

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.

368
Cchen_dev·2 days ago

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

331
Cchen_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.

194
Hhuang_ke·2 days ago

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

295
Kkernel_panic·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
Ttang_haoMod·28 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".

16
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.

75
Oops_wang·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.

259
Mmike_xu·1 hour agoedited

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

249
Rran_bo·3 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.

452
Sswoole_lee·1 hour ago

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

15
Cchen_dev·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.

356
Oops_wang·yesterdayedited

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

84
Bbob_chen·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.

149
Rran_bo·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.

340
Ddev_zhou·yesterday

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

478
Nnikic·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.

328
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.

20
Wwinter·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.

174
Kkite·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".

28
Ddev_zhou·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
Sslow_query·28 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.

10
Oops_wang·12 minutes ago

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

304
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.

1
Kkite·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.

92
Aalice_dev·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.

26
Ttang_hao·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.

11
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.

396
Aalice_devOP·2 days ago

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

88
Sslow_queryOP·12 minutes ago

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

169
Zzhu_zong·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".

110
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.

153
Hhuang_ke·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.

13
Mmike_xu·just now

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

12
KkiteOP·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.

1
Nnikic·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.

389
Mmike_xuOP·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
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.

18
Kkite·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.

191
Cchen_devOP·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.

95
Aalice_dev·2 days agoedited

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

1
Zzhu_zong·2 days ago

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

1
Oops_wang·28 minutes ago

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

224
Nnikic·2 days ago

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

223
Hhuang_keOP·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
Zzhou_yiMod·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.

403
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.

197
Hhuang_ke·2 days ago

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

208
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.

208
Kkite·2 days ago

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

187
Llinlin·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.

181
Rrase·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
Llinlin·2 days ago

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

168
Kkite·2 hours ago

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

150
Wwinter·28 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.

38
Ttang_hao·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.

174
Oops_wang·just now

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.

27
Hhuang_ke·28 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.

380
Mmike_xu·2 days agoLevel 6

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

348
Cchen_devOP·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.

153
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.

186
Rran_boOP·2 days agoLevel 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.

107
Cchen_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.

138
Ttang_hao·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.

134
Rran_bo·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.

117
Ddev_zhouOP·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.

39
Ddev_zhou·2 hours 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".

91
Ttang_hao·3 minutes 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.

9
Hhuang_ke·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.

70
Wwinter·1 hour ago

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

1
Hhuang_ke·1 hour 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
Sswoole_lee·3 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.

1
Sswoole_lee·2 days ago

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

1
Zzhou_yi·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.

75
Rran_bo·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.

75
Kkernel_panic·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
LlinlinMod·2 days ago

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

74
Bbob_chenOP·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.

9
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.

69
Hhuang_keOP·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.

1
Oops_wang·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.

49
Wwinter·2 days ago

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

381
Wwinter·just now

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

79
Lli_ming·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.

40
Kkite·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.

38
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.

116
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.

38
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.

30
Zzhou_yi·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.

29
Bbob_chen·2 days 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.

160
Rran_bo·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".

385
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.

166
Llinlin·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.

25
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.

24
Kkernel_panic·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.

14
Bbob_chen·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.

29
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.

1
Mmike_xu·12 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.

12
Kkernel_panic·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".

7
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.

5
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.

3
Ttang_hao·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.

3
Nnikic·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.

3
Rran_bo·2 days ago

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

3

This is the post detail page /en/c/polars-newbie/post/p11. 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 →