[usm] reliability, preparation for pool of objects, add Protocol.ReleaseStats() #32317
Conversation
yuri-lipnesh
added
changelog/no-changelog
team/usm
github-actions
bot
added
the
short review
Regression DetectorRegression Detector ResultsMetrics dashboard Baseline: 6fb76d5 Optimization Goals: ✅ No significant changes detected
|
| perf | experiment | goal | Δ mean % | Δ mean % CI | trials | links |
|---|---|---|---|---|---|---|
| ➖ | quality_gate_logs | % cpu utilization | +2.65 | [-0.34, +5.65] | 1 | Logs |
| ➖ | quality_gate_idle_all_features | memory utilization | +0.99 | [+0.84, +1.15] | 1 | Logs bounds checks dashboard |
| ➖ | uds_dogstatsd_to_api_cpu | % cpu utilization | +0.77 | [+0.05, +1.50] | 1 | Logs |
| ➖ | file_tree | memory utilization | +0.55 | [+0.39, +0.71] | 1 | Logs |
| ➖ | file_to_blackhole_1000ms_latency | egress throughput | +0.51 | [-0.27, +1.28] | 1 | Logs |
| ➖ | otel_to_otel_logs | ingress throughput | +0.18 | [-0.48, +0.84] | 1 | Logs |
| ➖ | file_to_blackhole_500ms_latency | egress throughput | +0.13 | [-0.64, +0.90] | 1 | Logs |
| ➖ | uds_dogstatsd_to_api | ingress throughput | +0.03 | [-0.08, +0.14] | 1 | Logs |
| ➖ | file_to_blackhole_100ms_latency | egress throughput | +0.00 | [-0.63, +0.64] | 1 | Logs |
| ➖ | tcp_dd_logs_filter_exclude | ingress throughput | +0.00 | [-0.01, +0.01] | 1 | Logs |
| ➖ | file_to_blackhole_300ms_latency | egress throughput | -0.02 | [-0.64, +0.61] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency_http1 | egress throughput | -0.02 | [-0.88, +0.84] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency_http2 | egress throughput | -0.02 | [-0.91, +0.86] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency | egress throughput | -0.08 | [-0.93, +0.78] | 1 | Logs |
| ➖ | file_to_blackhole_1000ms_latency_linear_load | egress throughput | -0.21 | [-0.66, +0.24] | 1 | Logs |
| ➖ | quality_gate_idle | memory utilization | -0.42 | [-0.48, -0.35] | 1 | Logs bounds checks dashboard |
| ➖ | tcp_syslog_to_blackhole | ingress throughput | -1.30 | [-1.37, -1.24] | 1 | Logs |
Bounds Checks: ❌ Failed
| perf | experiment | bounds_check_name | replicates_passed | links |
|---|---|---|---|---|
| ❌ | quality_gate_idle | memory_usage | 8/10 | bounds checks dashboard |
| ✅ | file_to_blackhole_0ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http1 | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http1 | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http2 | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http2 | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_1000ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_1000ms_latency_linear_load | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_100ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_100ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_300ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_300ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_500ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_500ms_latency | memory_usage | 10/10 | |
| ✅ | quality_gate_idle_all_features | memory_usage | 10/10 | bounds checks dashboard |
| ✅ | quality_gate_logs | lost_bytes | 10/10 | |
| ✅ | quality_gate_logs | memory_usage | 10/10 |
Explanation
Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%
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
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:
-
Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.
-
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.
-
Its configuration does not mark it "erratic".
CI Pass/Fail Decision
❌ Failed. Some Quality Gates were violated.
- quality_gate_idle, bounds check memory_usage: 8/10 replicas passed. Failed 2 which is > 0. Gate FAILED.
- quality_gate_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check lost_bytes: 10/10 replicas passed. Gate passed.
- quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
yuri-lipnesh
force-pushed
the
yuri.lipnesh/USMON-1324_release_protocol_stats
branch
from
ad6f1d4 to
45fe447
Compare
github-actions
bot
added
medium review
Test changes on VMUse this command from test-infra-definitions to manually test this PR changes on a VM: inv aws.create-vm --pipeline-id=51669806 --os-family=ubuntuNote: This applies to commit 196aaf9 |
Uncompressed package size comparisonComparison with ancestor Diff per package
Decision |
github-actions
bot
added
long review
…tats.ReleaseStats()
yuri-lipnesh
changed the title
|
|
||
| nt.tracer.ReleaseProtocolStats(cs) |
There was a problem hiding this comment.
why is it here? why do we need to expose a method in the NPM tracer to release USM stats?
There was a problem hiding this comment.
Tracer.Connections keeps reference to stats objects. The eventual purpose is to release objects to pool, it could be done only after Tracer no longer references stats objects.
| // CanRelease is true when the struct is no longer referenced | ||
| CanRelease bool |
There was a problem hiding this comment.
the current design frees the memory only if we called CombineWith, but there are other flows in which we don't call combineWith but still must release the resources
There was a problem hiding this comment.
The goal is to introduce changes gradually, in stages, with the remaining cases being considered later, in the following stages.
| // releaseProtocolStats release stats objects for enabled protocols. | ||
| func (e *ebpfProgram) releaseProtocolStats(conns *network.Connections) { | ||
| for _, protocol := range e.enabledProtocols { | ||
| switch protocol.Instance.Type() { | ||
| case protocols.HTTP: | ||
| for _, rs := range conns.HTTP { | ||
| rs.ReleaseStats() | ||
| } | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
Having USM main entrypoint getting NPM representation of connections, just to free internal state is not the right way to do it.
There was a problem hiding this comment.
I intend to change this in the next phase, this approach allowed me to keep the changes at this stage minimal.
pkg/network/tracer/tracer.go
Outdated
| t.bufferLock.Lock() | ||
| defer t.bufferLock.Unlock() |
There was a problem hiding this comment.
for precautionary reasons because it is used in Tracer.GetActiveConnections, however this lock here seems to be redundant. I'll remove it and check the impact.
….Lock() in Tracer.ReleaseProtocolStats()
…ineWith(), each newRequests.CanRelease=true
What does this PR do?
It adds usm Protocol.ReleaseStats() to release reported stats objects.
It introduces a chain of functions to invoke ReleaseStats(), starting from the network tracer.
The HTTP handler retains and implicitly releases references to the reported statistics.
Motivation
This is part of the ongoing reliability effort, serving as a preparation step for adding a pool of DDSketch objects.
It breaks the task of implementing the object pool into several distinct phases.
Describe how you validated your changes
Unit tests.
Running usm http load tests.
Possible Drawbacks / Trade-offs
Additional Notes