Reduce GlobalObjectStore contention by sampling underlying map size

[mcculls]

What Does This Do

Reduce GlobalObjectStore contention by sampling underlying map size periodically, rather on every write

Motivation

ConcurrentHashMap.size() must sum counts across the different buckets which - we don't need to do this on every write, instead we track how many times the thread missed lookups and fell-through to use a store key. We then use this to decide when to sample the map size. This is sufficient because the thresholds are several times the sample rate.

We also update the sampled size on every scheduled cleanup.

Contributor Checklist

Jira ticket: [PROJ-IDENT]

[mcculls]

@DataDog review

[mcculls]

@DataDog review

[amarziali]

nice. lgtm I just left a couple of informative comments

[Copilot]

🟢 Approval recommended

The functional changes are cohesive and localized, with only minor documentation clarifications suggested.

Review details

Suppressed comments (1)

field-inject/src/main/java/datadog/instrument/fieldinject/GlobalObjectStore.java:350

• LookupKey.misses is incremented by both miss() and skip(), so describing it as only "lookup missed" is inaccurate; clarifying this helps prevent confusion about what drives map-size sampling.

    /** Number of times a lookup missed when writing. */
    int misses;

• Files reviewed: 1/1 changed files
• Comments generated: 1
• Review effort level: Lite

[datadog-datadog-prod-us1-2]

More details

The PR reduces map-size checks during writes. The latest commit makes each thread sample on its first write, so it fixes the earlier unbounded-growth case for short-lived threads; no open findings remain.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Datadog Autotest · Commit d270f39 · What is Autotest? · @DataDog review to ask questions · Any feedback? Reach out in #autotest}