Widen CompiledMetric tag-cache key from 32 to 57 bits

[gmalette]

Situation

CompiledMetric keys its tag-combination cache by a 32-bit rotate-left-5 + XOR hash of the tag values. Collisions are detected by comparing full tag values but never resolved: the losing tag combination is never cached, so every emit of it for the rest of the process lifetime allocates a fresh PrecompiledDatagram (the allocation CompiledMetric exists to avoid) and increments statsd_instrument.compiled_metric.hash_collision_detected. Once a colliding pair lands among hot tag combinations, the counter fires continuously until the process restarts. Because String#hash is seeded per process, which pair collides differs per process, so the signal shows up as intermittent, fleet-wide noise.

At 32 bits this is inevitable at real volumes. A process that has cached D distinct tag combinations sees expected collisions of roughly D² / 2³³, and every new distinct combination looked up against a full 5000-entry cache collides with probability ~1.2e-6. A busy fleet produces a steady stream of collision events every day.

Execution

Widen the key to 57 bits at zero cost. The only operation that can escape Fixnum range is the intermediate (__cache_key__ << 5), and 57-bit keys keep it within 2**62 - 1 on 64-bit CRuby:

k <= 2**57 - 1
k << 5 <= 2**62 - 32
(k << 5) | (k >> 52) <= 2**62 - 1

String#hash exposes ~60 bits of SipHash entropy, so the extra 25 bits are real. Expected collisions per process drop by 2^25 (~33 million-fold), turning a daily occurrence into a never-in-practice one, with the same op count and no new allocations on either the hit or miss path. The collision-detection branch stays as a correctness guard. The new test derives a low-32-bit-colliding pair at runtime (hash seeds are process-local) and asserts both combinations cache independently with no collision metric.

This touches the same lines as #413 but is orthogonal: that changes the mixing function for speed, this widens the key space. Either can land first; the conflict is trivial to resolve.

[Copilot]

🟡 Not ready to approve

The newly added tests can hang or fail across the repository’s supported Ruby engines (CI includes JRuby/TruffleRuby) due to unbounded collision-search logic and non-portable GC allocation stats usage.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

This review doesn't count toward merge requirements. Sign up for the private preview to control whether Copilot approvals count.

Pull request overview

This PR widens CompiledMetric’s tag-combination cache key from a 32-bit masked rotate-left+XOR hash to a 57-bit masked variant to drastically reduce practical collision rates while keeping integer operations within the immediate/small-integer range on 64-bit CRuby.

Changes:

• Expand the cache-key mask to 57 bits and adjust rotation to match the wider key space.
• Add tests intended to validate wider-than-32-bit behavior and guard against allocation regressions in cache-key computation.
• Bump library version to 3.11.2 and document the change in the changelog.

File summaries

File	Description
lib/statsd/instrument/compiled_metric.rb	Widens tag-cache key mask to 57 bits and updates the rotation shift accordingly.
test/compiled_metric_test.rb	Updates collision test mask and adds new tests for 57-bit key behavior and allocation characteristics.
lib/statsd/instrument/version.rb	Bumps gem version to 3.11.2.
CHANGELOG.md	Adds a 3.11.2 entry documenting the widened cache key.

Review details

Suppressed comments (1)

test/compiled_metric_test.rb:333

• This collision-search loop is unbounded and currently uses i.hash, which can make the test extremely slow or non-terminating on some Rubies. Use String#hash (seeded per process) and add a max-iteration guard so the test fails fast instead of hanging CI.

    until pair
      key = i.hash & 0xFFFFFFFF
      other = seen[key]
      if other && other != i && (other.hash & 0x01FFFFFFFFFFFFFF) != (i.hash & 0x01FFFFFFFFFFFFFF)
        pair = [other, i]

• Files reviewed: 4/4 changed files
• Comments generated: 2
• Review effort level: Lite

We're testing this review assessment. Please use 👍 or 👎 to tell us if it's correct.