Bound evaluation queue memory

[leoromanovsky]

Motivation

The 4,096-entry evaluation queue limits event count, but it does not limit retained context bytes. JOL measured 81.3 MB for a full queue of flattened 100-field contexts. A full queue of maximum legal contexts projects 672.1 MB. Slow EVP delivery can retain this memory.

This change limits the conservative queue estimate to 16 MiB.

Changes and Decisions

• Calculate context bytes during the existing bounded context copy. Enqueue and release do not walk the context.
• Store the immutable total on each event. This avoids a second queue wrapper allocation.
• Reserve bytes with a non-blocking atomic operation before the queue offer. Keep 4,096 entries as a secondary limit.
• Batch worker releases to 64 KiB. This reduces producer and worker cache-line handoff.
• Release all reservations during drain and close. Use a separate atomic path for producer cancellation.
• Report queue_byte_budget separately from queue_overflow.

Validation

• JOL: the 100-field profile stopped at 639 events and retained 12.1 MiB. The maximum profile stopped at 56 events and retained 8.8 MiB.
• JMH: writer enqueue increased from 15.0-16.4 ns to 17.7-19.5 ns across three profiles. Allocation increased by 8 bytes per event.
• JMH: the 100-field hook increased from 1,044.4 ns to 1,086.9 ns, or 4.1%. Allocation increased by 8 bytes.
• JMH producer/consumer groups: one producer increased from 53.1 ns to 77.4 ns. Four and 16 producer results had overlapping confidence intervals.

These are controlled JDK 26 measurements. They are not production results.

[datadog-datadog-us1-prod]

🎯 Code Coverage (details)
• Patch Coverage: 95.28%
• Overall Coverage: 78.32% (+19.97%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: be0d007 | Docs | Datadog PR Page | Give us feedback!}

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	14.01 s	13.96 s	[-0.4%; +1.1%] (no difference)
startup:insecure-bank:tracing:Agent	12.98 s	12.99 s	[-1.1%; +1.0%] (no difference)
startup:petclinic:appsec:Agent	17.42 s	16.68 s	[+0.0%; +8.8%] (maybe worse)
startup:petclinic:iast:Agent	16.76 s	17.41 s	[-8.0%; +0.5%] (no difference)
startup:petclinic:profiling:Agent	17.31 s	17.23 s	[-0.8%; +1.7%] (no difference)
startup:petclinic:sca:Agent	17.10 s	17.20 s	[-1.8%; +0.6%] (no difference)
startup:petclinic:tracing:Agent	16.41 s	16.49 s	[-1.4%; +0.4%] (no difference)

Commit: be0d0078 · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.