ECEdge Computing · Resources
Community homeHot discussions Latest posts Best of Knowledge base Resources About

Edge Computing · Resources · Building an internal platform with edge-computing-resources from scratch: architecture and decision log

1.9K
ECr/edge-computing-resources·posted by huang_ke·3 hours agoHelp

Building an internal platform with edge-computing-resources from scratch: architecture and decision log

Short version: edge-computing-resources 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.

# 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

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.

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.

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.

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

431 comments

431 comments

· first 120 loaded
M
Rran_bo·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.

486
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.

281
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.

464
Lli_ming·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.

449
Mmike_xu·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.

422
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.

342
Kkernel_panic·2 days ago

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

318
Nnikic·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.

300
Oops_wang·yesterday

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

181
Sswoole_leeOP·2 days ago

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

199
Nnikic·28 minutes agoedited

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

118
Oops_wang·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.

389
Zzhu_zongOP·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.

2
Lli_ming·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.

5
Aalice_devMod·2 days ago

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

12
Rrase·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.

81
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.

1
Mmike_xu·5 hours ago

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

4
Ttang_hao·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.

330
Sslow_query·2 days ago

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

459
Zzhou_yi·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.

301
Cchen_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.

140
Kkernel_panicMod·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.

144
Aalice_devOP·2 days agoLevel 6

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

494
Ddev_zhou·2 days agoLevel 6

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

428
Llinlin·2 days agoLevel 6

One counter-example: below edge-computing-resources 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
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.

125
Oops_wang·2 days ago

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

19
Sslow_query·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.

217
Zzhu_zong·yesterdayLevel 6

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

18
Nnikic·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.

132
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.

260
Lli_ming·2 days agoLevel 6

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

344
Ttang_hao·28 minutes agoedited

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

34
Nnikic·yesterdayLevel 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
Zzhu_zong·28 minutes 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.

2
Llinlin·yesterday

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.

120
Zzhou_yiOP·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.

9
Oops_wang·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.

132
Rrase·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
Kkite·2 days ago

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

234
Sslow_query·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.

213
Lli_ming·2 days 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.

447
Sswoole_lee·28 minutes ago

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

101
Sswoole_lee·28 minutes ago

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

285
Rrase·2 days ago

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

275
Sswoole_leeMod·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.

280
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.

261
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.

88
Lli_ming·2 days 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.

185
Rran_bo·2 days ago

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

178
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.

8
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.

258
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.

250
Zzhu_zong·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.

165
KkiteMod·2 days ago

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

143
Bbob_chen·2 days ago

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

120
Sslow_query·2 days ago

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

1
Oops_wang·2 days ago

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

3
NnikicOP·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.

29
Llinlin·just now

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

80
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.

72
Sswoole_lee·2 days ago

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

58
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.

12
Zzhu_zong·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.

68
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.

448
Oops_wang·2 days agoedited

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

119
Sswoole_lee·just now

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

65
Nnikic·2 days ago

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

117
Aalice_dev·2 hours ago

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

344
Nnikic·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.

21
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.

481
Kkernel_panic·2 days ago

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

194
Sslow_query·2 days agoLevel 6

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

475
Zzhu_zong·2 days agoLevel 6

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

383
Nnikic·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.

347
Sslow_query·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.

320
Bbob_chen·2 days ago

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

7
Lli_ming·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.

400
Kkernel_panic·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.

107
Sswoole_lee·2 days agoedited

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

4
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.

162
Oops_wang·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.

22
Rran_bo·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.

7
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.

2
Sswoole_lee·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.

118
Llinlin·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.

28
Aalice_dev·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.

1
KkiteOP·2 days ago

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

47
Aalice_dev·2 days ago

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

99
Zzhu_zong·2 days agoedited

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

316
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.

65
NnikicOP·28 minutes ago

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

1
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.

203
Bbob_chen·2 days ago

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

10
Mmike_xu·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.

425
Cchen_dev·3 minutes ago

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

14
Bbob_chen·2 days ago

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

104
Rran_bo·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.

1
Zzhu_zong·2 days ago

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

61
Aalice_dev·2 hours ago

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

58
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.

42
Hhuang_ke·2 days ago

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

41
Bbob_chen·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.

404
Ddev_zhou·2 days ago

One counter-example: below edge-computing-resources 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
Oops_wang·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.

39
Mmike_xu·12 minutes ago

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

26
Bbob_chen·2 days ago

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

277
Wwinter·2 days ago

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

24
Wwinter·2 days ago

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

22
Bbob_chen·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.

16
Mmike_xu·2 days 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.

15
Sswoole_lee·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.

11
Bbob_chen·2 days agoedited

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

8
Rrase·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.

6
Bbob_chen·12 minutes ago

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

4
Rrase·2 days ago

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

1
Ddev_zhou·2 days ago

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

3
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.

2
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.

1

This is the post detail page /en/c/edge-computing-resources/post/p12. 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 →