Add "no_wait" mode for file synchronization

[haze518]

This PR introduces a new "no_wait" mode to the file synchronization process. The "nowait" mode allows file operations to proceed without waiting for certain synchronization steps to complete, improving performance in scenarios where blocking is unnecessary

[coveralls]

Pull Request Test Coverage Report for Build 12687277738

Details

• 214 of 243 (88.07%) changed or added relevant lines in 24 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage increased (+0.05%) to 75.37%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
server/src/streaming/streams/storage.rs	2	4	50.0%
server/src/streaming/topics/storage.rs	2	4	50.0%
server/src/streaming/storage.rs	5	8	62.5%
server/src/streaming/segments/storage.rs	7	14	50.0%
server/src/streaming/persistence/task.rs	63	78	80.77%

Files with Coverage Reduction	New Missed Lines	%
server/src/configs/system.rs	1	85.0%

Totals
Change from base Build 12639344475:	0.05%
Covered Lines:	24343
Relevant Lines:	32298

---

💛 - Coveralls

[spetz]

Great job, I can certainly see the improvements with nowait mode (as expected). I've just executed a quick few tests on M1 Max, and here are the example results for the following benchmark:

cargo r --bin iggy-bench -r send --streams 10 --producers 10 --message-size 1000 --messages-per-batch 1000 --message-batches 3000 tcp

nowait

Results: Total throughput: 1966.74 MB/s, 1920640 messages/s, average throughput: 196.67 MB/s, average p50 latency: 4.22 ms, average p90 latency: 5.13 ms, average p95 latency: 5.48 ms, average p99 latency: 6.83 ms, average p999 latency: 255.91 ms, average latency: 5.21 ms, median latency: 4.22 ms

wait

Results: Total throughput: 1671.96 MB/s, 1632774 messages/s, average throughput: 167.20 MB/s, average p50 latency: 3.77 ms, average p90 latency: 4.99 ms, average p95 latency: 6.06 ms, average p99 latency: 20.11 ms, average p999 latency: 589.00 ms, average latency: 6.13 ms, median latency: 3.77 ms

I will do a few more tests tomorrow on Linux PC, but it looks like it could be merged already :)

[spetz]

Looks good, and works even better :) Please update the Cargo.toml in server to 0.4.100 so that it will be automatically released.

[spetz]

And a quick update after running the same benchmark for nowait mode on Linux PC (7950X, 64GB RAM, Gen 4 NVMe).

Results: Total throughput: 3212.28 MB/s, 3136991 messages/s, average throughput: 321.23 MB/s, average p50 latency: 2.66 ms, average p90 latency: 5.14 ms, average p95 latency: 6.22 ms, average p99 latency: 8.73 ms, average p999 latency: 12.16 ms, average latency: 3.19 ms, median latency: 2.66 ms

[hubcio]

LGTM

[hubcio]

like promised, some charts, master:

no_wait:

[haze518]

like promised, some charts, master:

no_wait:

Yes, now I can see the difference :) When I was testing, I was focusing on the median and the message throughput rather than the 90+ percentiles. Thanks!