Use poll_recv_many for processing commands to be written to the nats server #1189

paolobarbolini · 2024-01-04T08:41:16Z

Tokio 1.36 exposed Receiver::poll_recv_many, which allows polling multiple items at once from an mpsc channel. Considering that our write buffer already has a lenient capacity limit, and the benchmarks show that the reduced overhead is worth it, this PR makes the connection handler poll 16 commands at once instead of just 1.

Benchmark results (Ryzen 5900X)

Gnuplot not found, using plotters backend
nats::publish_throughput/32
                        time:   [86.357 ms 91.881 ms 95.353 ms]
                        thrpt:  [160.02 MiB/s 166.07 MiB/s 176.69 MiB/s]
                 change:
                        time:   [-27.813% -21.773% -15.602%] (p = 0.00 < 0.05)
                        thrpt:  [+18.487% +27.834% +38.529%]
                        Performance has improved.
nats::publish_throughput/1024
                        time:   [170.67 ms 183.75 ms 197.30 ms]
                        thrpt:  [2.4168 GiB/s 2.5951 GiB/s 2.7939 GiB/s]
                 change:
                        time:   [-30.036% -23.320% -15.435%] (p = 0.00 < 0.05)
                        thrpt:  [+18.253% +30.412% +42.931%]
                        Performance has improved.
Benchmarking nats::publish_throughput/8192: Warming up for 1.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 10.8s.
nats::publish_throughput/8192
                        time:   [1.0787 s 1.1174 s 1.1481 s]
                        thrpt:  [3.3227 GiB/s 3.4139 GiB/s 3.5363 GiB/s]
                 change:
                        time:   [-16.965% -12.838% -8.4008%] (p = 0.00 < 0.05)
                        thrpt:  [+9.1712% +14.729% +20.431%]
                        Performance has improved.
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) low mild

Benchmarking nats::publish_amount/32: Warming up for 1.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 5.4s or enable flat sampling.
nats::publish_amount/32 time:   [77.440 ms 88.434 ms 96.639 ms]
                        thrpt:  [5.1739 Melem/s 5.6539 Melem/s 6.4566 Melem/s]
                 change:
                        time:   [-35.253% -26.741% -17.500%] (p = 0.00 < 0.05)
                        thrpt:  [+21.212% +36.502% +54.447%]
                        Performance has improved.
nats::publish_amount/1024
                        time:   [163.45 ms 174.96 ms 187.47 ms]
                        thrpt:  [2.6670 Melem/s 2.8578 Melem/s 3.0590 Melem/s]
                 change:
                        time:   [-35.579% -29.700% -23.342%] (p = 0.00 < 0.05)
                        thrpt:  [+30.449% +42.248% +55.229%]
                        Performance has improved.
Benchmarking nats::publish_amount/8192: Warming up for 1.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 11.3s.
nats::publish_amount/8192
                        time:   [1.0952 s 1.1479 s 1.1899 s]
                        thrpt:  [420.19 Kelem/s 435.59 Kelem/s 456.55 Kelem/s]
                 change:
                        time:   [-16.233% -11.968% -7.6855%] (p = 0.00 < 0.05)
                        thrpt:  [+8.3254% +13.594% +19.379%]
                        Performance has improved.
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) low mild

nats::subscribe_amount/32
                        time:   [279.04 ms 323.40 ms 363.97 ms]
                        thrpt:  [1.3737 Melem/s 1.5461 Melem/s 1.7919 Melem/s]
                 change:
                        time:   [-26.028% -12.368% +2.3029%] (p = 0.14 > 0.05)
                        thrpt:  [-2.2511% +14.114% +35.186%]
                        No change in performance detected.
nats::subscribe_amount/1024
                        time:   [356.16 ms 373.04 ms 389.21 ms]
                        thrpt:  [1.2846 Melem/s 1.3404 Melem/s 1.4038 Melem/s]
                 change:
                        time:   [-10.456% -4.4544% +1.6193%] (p = 0.20 > 0.05)
                        thrpt:  [-1.5935% +4.6621% +11.677%]
                        No change in performance detected.
Benchmarking nats::subscribe_amount/8192: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 16.0s.
nats::subscribe_amount/8192
                        time:   [1.5499 s 1.5847 s 1.6154 s]
                        thrpt:  [309.52 Kelem/s 315.52 Kelem/s 322.61 Kelem/s]
                 change:
                        time:   [-4.5167% -1.9550% +0.2916%] (p = 0.16 > 0.05)
                        thrpt:  [-0.2907% +1.9940% +4.7304%]
                        No change in performance detected.
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) low mild

Benchmarking nats::request_amount/32: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 7.7s.
nats::request_amount/32 time:   [767.62 ms 783.32 ms 798.11 ms]
                        thrpt:  [12.530 Kelem/s 12.766 Kelem/s 13.027 Kelem/s]
                 change:
                        time:   [-4.2668% -0.2408% +3.9516%] (p = 0.91 > 0.05)
                        thrpt:  [-3.8014% +0.2414% +4.4569%]
                        No change in performance detected.
Benchmarking nats::request_amount/1024: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 7.4s.
nats::request_amount/1024
                        time:   [789.21 ms 802.72 ms 817.36 ms]
                        thrpt:  [12.234 Kelem/s 12.458 Kelem/s 12.671 Kelem/s]
                 change:
                        time:   [-4.8586% -2.2998% +0.4285%] (p = 0.12 > 0.05)
                        thrpt:  [-0.4267% +2.3539% +5.1067%]
                        No change in performance detected.
Benchmarking nats::request_amount/8192: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 8.6s.
nats::request_amount/8192
                        time:   [843.46 ms 858.24 ms 872.76 ms]
                        thrpt:  [11.458 Kelem/s 11.652 Kelem/s 11.856 Kelem/s]
                 change:
                        time:   [-2.2396% +1.3327% +5.8574%] (p = 0.57 > 0.05)
                        thrpt:  [-5.5333% -1.3151% +2.2909%]
                        No change in performance detected.

Jarema · 2024-01-04T10:58:13Z

That is really nice improvement @paolobarbolini !
Is there timeline when it will be released?

paolobarbolini · 2024-01-04T11:06:46Z

From what I can gather from the latest releases it shouldn't take more than a few weeks.
This is what I got from the tokio-rs Discord 😄 https://discord.com/channels/500028886025895936/500336346770964480/1192423210780737578

paolobarbolini · 2024-01-30T15:18:58Z

Tokio v1.36 is out

paolobarbolini · 2024-01-30T15:21:06Z

Tokio v1.36 is out

Ok it's not. Just a PR 😞
tokio-rs/tokio#6312

paolobarbolini · 2024-02-02T13:08:01Z

Now it's out for real

Jarema

This is awesome to leverage this new API and gain 30% performance!

Thanks @paolobarbolini

Just some minor questions.

async-nats/src/lib.rs

Jarema

LGTM!

paolobarbolini force-pushed the poll-recv-many branch from 88bda9c to 46ffeda Compare January 4, 2024 10:47

paolobarbolini force-pushed the poll-recv-many branch from 46ffeda to 8b11a0b Compare January 30, 2024 15:18

paolobarbolini marked this pull request as ready for review January 30, 2024 15:18

paolobarbolini force-pushed the poll-recv-many branch from 8b11a0b to 0489470 Compare January 30, 2024 15:21

Use poll_recv_many for receiving messages to publish to the connection

45df788

paolobarbolini force-pushed the poll-recv-many branch from 0489470 to 45df788 Compare February 2, 2024 13:07

Jarema reviewed Feb 2, 2024

View reviewed changes

async-nats/src/lib.rs Show resolved Hide resolved

async-nats/src/lib.rs Show resolved Hide resolved

Jarema approved these changes Feb 2, 2024

View reviewed changes

Jarema merged commit 00f1d56 into nats-io:main Feb 2, 2024
11 checks passed

Jarema mentioned this pull request Mar 4, 2024

Release async-nats/v0.34.0 #1228

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use poll_recv_many for processing commands to be written to the nats server #1189

Use poll_recv_many for processing commands to be written to the nats server #1189

paolobarbolini commented Jan 4, 2024 •

edited

Loading

Jarema commented Jan 4, 2024

paolobarbolini commented Jan 4, 2024

paolobarbolini commented Jan 30, 2024

paolobarbolini commented Jan 30, 2024

paolobarbolini commented Feb 2, 2024

Jarema left a comment

Jarema left a comment

Use poll_recv_many for processing commands to be written to the nats server #1189

Use poll_recv_many for processing commands to be written to the nats server #1189

Conversation

paolobarbolini commented Jan 4, 2024 • edited Loading

Jarema commented Jan 4, 2024

paolobarbolini commented Jan 4, 2024

paolobarbolini commented Jan 30, 2024

paolobarbolini commented Jan 30, 2024

paolobarbolini commented Feb 2, 2024

Jarema left a comment

Choose a reason for hiding this comment

Jarema left a comment

Choose a reason for hiding this comment

paolobarbolini commented Jan 4, 2024 •

edited

Loading