fix(compression): Fix gzip and zlib performance degradation

[Hexta]

Fix gzip and zlib performance degradation caused by:

• flate2 v1.0.28 started to resize its input buffer up to its capacity and back to the actual number of bytes written.
• Some sinks are writing to Compressor without buffering, resulting in frequent small writes to the flate2 writer. Within 32KB of input buffer in flate2, this causes an excessive number of memset operations and degraded sink throughput.

This fix introduces a wrapper buffer in front of gzip and zlib writers to accumulate data before calling the write function of the underlying writer.

fixes: #19981

[pront]

/ci-run-regression

[jszwedko]

/ci-run-regression

Unfortunately this doesn't work on forks 😢 See #18437

[Hexta]

Only gzip and zlib compressors gained performance improvements with the input buffer.
I decided not to use this for other compressors for now.

[github-actions]

Regression Detector Results

Run ID: d1c972eb-b8ef-4138-8334-731988159bc2
Baseline: 8db6288
Comparison: ecdcba8
Total CPUs: 7

Performance changes are noted in the perf column of each table:

• ✅ = significantly better comparison variant performance
• ❌ = significantly worse comparison variant performance
• ➖ = no significant change in performance

No significant changes in experiment optimization goals

Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%

There were no significant changes in experiment optimization goals at this confidence level and effect size tolerance.

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI
➖	otlp_http_to_blackhole	ingress throughput	+4.27	[+4.11, +4.43]
➖	socket_to_socket_blackhole	ingress throughput	+3.63	[+3.56, +3.70]
➖	splunk_hec_route_s3	ingress throughput	+3.04	[+2.53, +3.55]
➖	datadog_agent_remap_blackhole	ingress throughput	+2.03	[+1.93, +2.12]
➖	syslog_log2metric_humio_metrics	ingress throughput	+1.68	[+1.56, +1.80]
➖	syslog_humio_logs	ingress throughput	+1.18	[+1.07, +1.29]
➖	http_to_http_acks	ingress throughput	+1.16	[-0.16, +2.49]
➖	datadog_agent_remap_blackhole_acks	ingress throughput	+1.03	[+0.93, +1.13]
➖	syslog_log2metric_tag_cardinality_limit_blackhole	ingress throughput	+0.87	[+0.76, +0.98]
➖	datadog_agent_remap_datadog_logs_acks	ingress throughput	+0.81	[+0.73, +0.89]
➖	file_to_blackhole	egress throughput	+0.42	[-2.06, +2.91]
➖	datadog_agent_remap_datadog_logs	ingress throughput	+0.42	[+0.34, +0.50]
➖	http_to_s3	ingress throughput	+0.31	[+0.03, +0.59]
➖	http_text_to_http_json	ingress throughput	+0.15	[+0.03, +0.26]
➖	http_to_http_json	ingress throughput	+0.03	[-0.04, +0.11]
➖	http_to_http_noack	ingress throughput	+0.03	[-0.07, +0.12]
➖	splunk_hec_to_splunk_hec_logs_acks	ingress throughput	+0.00	[-0.15, +0.15]
➖	splunk_hec_indexer_ack_blackhole	ingress throughput	-0.00	[-0.14, +0.14]
➖	http_elasticsearch	ingress throughput	-0.01	[-0.09, +0.07]
➖	splunk_hec_to_splunk_hec_logs_noack	ingress throughput	-0.04	[-0.16, +0.07]
➖	otlp_grpc_to_blackhole	ingress throughput	-0.09	[-0.18, -0.01]
➖	enterprise_http_to_http	ingress throughput	-0.11	[-0.19, -0.03]
➖	syslog_loki	ingress throughput	-0.41	[-0.46, -0.36]
➖	fluent_elasticsearch	ingress throughput	-0.74	[-1.21, -0.27]
➖	syslog_splunk_hec_logs	ingress throughput	-0.98	[-1.05, -0.91]
➖	syslog_regex_logs2metric_ddmetrics	ingress throughput	-1.11	[-1.20, -1.03]
➖	syslog_log2metric_splunk_hec_metrics	ingress throughput	-1.43	[-1.57, -1.30]

Explanation

A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".

For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.

3. Its configuration does not mark it "erratic".

[bruceg]

Thanks for this, @Hexta. The code looks good to me.

[bruceg]

Did you happen to do any testing in the original scenario with different buffer sizes?

[Hexta]

Yes, I tried both 2k and 8k.
2k: throughput reduced by 1.4%.
8k: no changes.

[github-actions]

Regression Detector Results

Run ID: 9d93c793-ba4b-4588-a124-39700197957c
Baseline: 485dea7
Comparison: d07a435
Total CPUs: 7

Performance changes are noted in the perf column of each table:

• ✅ = significantly better comparison variant performance
• ❌ = significantly worse comparison variant performance
• ➖ = no significant change in performance

No significant changes in experiment optimization goals

Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%

There were no significant changes in experiment optimization goals at this confidence level and effect size tolerance.

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI
➖	syslog_regex_logs2metric_ddmetrics	ingress throughput	+2.59	[+2.44, +2.73]
➖	syslog_log2metric_splunk_hec_metrics	ingress throughput	+2.48	[+2.32, +2.63]
➖	otlp_http_to_blackhole	ingress throughput	+1.71	[+1.54, +1.88]
➖	syslog_splunk_hec_logs	ingress throughput	+0.78	[+0.73, +0.84]
➖	syslog_humio_logs	ingress throughput	+0.59	[+0.48, +0.70]
➖	fluent_elasticsearch	ingress throughput	+0.58	[+0.11, +1.05]
➖	syslog_log2metric_humio_metrics	ingress throughput	+0.55	[+0.39, +0.71]
➖	syslog_loki	ingress throughput	+0.50	[+0.43, +0.57]
➖	syslog_log2metric_tag_cardinality_limit_blackhole	ingress throughput	+0.37	[+0.26, +0.48]
➖	http_to_http_noack	ingress throughput	+0.20	[+0.11, +0.29]
➖	http_to_s3	ingress throughput	+0.15	[-0.13, +0.43]
➖	http_text_to_http_json	ingress throughput	+0.07	[-0.05, +0.20]
➖	http_to_http_json	ingress throughput	+0.02	[-0.05, +0.09]
➖	splunk_hec_indexer_ack_blackhole	ingress throughput	+0.01	[-0.13, +0.15]
➖	splunk_hec_to_splunk_hec_logs_acks	ingress throughput	+0.00	[-0.15, +0.15]
➖	splunk_hec_route_s3	ingress throughput	-0.05	[-0.55, +0.46]
➖	splunk_hec_to_splunk_hec_logs_noack	ingress throughput	-0.05	[-0.16, +0.07]
➖	enterprise_http_to_http	ingress throughput	-0.08	[-0.16, -0.01]
➖	http_elasticsearch	ingress throughput	-0.11	[-0.17, -0.04]
➖	datadog_agent_remap_datadog_logs_acks	ingress throughput	-0.28	[-0.35, -0.20]
➖	datadog_agent_remap_datadog_logs	ingress throughput	-0.28	[-0.39, -0.18]
➖	otlp_grpc_to_blackhole	ingress throughput	-0.29	[-0.38, -0.20]
➖	datadog_agent_remap_blackhole	ingress throughput	-0.33	[-0.43, -0.24]
➖	datadog_agent_remap_blackhole_acks	ingress throughput	-0.43	[-0.53, -0.33]
➖	file_to_blackhole	egress throughput	-0.75	[-3.12, +1.62]
➖	http_to_http_acks	ingress throughput	-1.30	[-2.61, +0.02]
➖	socket_to_socket_blackhole	ingress throughput	-1.58	[-1.64, -1.51]

Explanation

A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".

For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.

3. Its configuration does not mark it "erratic".