feat(init): add cold-start init-phase timing instrumentation

[duncanista]

Please include Jira ticket in title. (no ticket yet — add your SLS ticket before marking ready for review.)

Overview

Adds debug-gated init-phase timing instrumentation to the extension entry point so cold-start (initialization) time can be attributed per phase and measured rather than guessed. This is the measurement prerequisite (H0) for the cold-start improvement effort.

A small helper log_init_checkpoint(start_time, phase) is called at each key boundary in main() / extension_loop_active() and logs both:

• delta — time since the previous checkpoint (that phase's own cost), and
• cumulative — time since process start,

in milliseconds to 6 decimal places (nanosecond resolution), so sub-millisecond phases are visible and per-phase cost is read directly off each line — no subtracting consecutive values by hand. Phases:

crypto_provider_ready → tls_client_build → config_parse → shared_client_ready → register_ready → dogstatsd_started → trace_agent_started → telemetry_subscribed → ready

These isolate the costs the cold-start work targets — e.g. tls_client_build and shared_client_ready are the two redundant reqwest/TLS client builds (H1), register_ready/telemetry_subscribed are the two Extensions-API round-trips (H2). It also logs available_parallelism() once — the value Tokio sizes its worker pool from, which varies per Lambda memory tier (H15).

Example output (only at DD_LOG_LEVEL=debug):

INIT | available_parallelism=Ok(2)
INIT | phase=crypto_provider_ready delta=0.412000ms cumulative=0.412000ms
INIT | phase=tls_client_build delta=5.931200ms cumulative=6.343200ms
INIT | phase=config_parse delta=1.204050ms cumulative=7.547250ms
INIT | phase=shared_client_ready delta=4.883100ms cumulative=12.430350ms
INIT | phase=register_ready delta=11.077000ms cumulative=23.507350ms
...
INIT | phase=ready delta=2.918000ms cumulative=41.902000ms

Notes

• Effectively free at the default info level: the per-phase bookkeeping (clock read + atomic) is guarded behind a DEBUG-level check and returns early otherwise; nothing is logged or computed.
• Nanosecond-resolution timing underneath; per-phase delta is tracked across the two functions via a module-level AtomicU64 of nanos (checkpoints run sequentially on the init path; saturating_sub guards against any reordering).
• No behavior change — debug-only observability. grep 'INIT |' yields the full timeline as one series.

Testing

• cargo clippy --bin bottlecap — clean (crate denies clippy::all + clippy::pedantic + clippy::unwrap_used).
• cargo fmt — clean.
• Manual (to do): run with DD_LOG_LEVEL=debug via the local_tests/ RIE harness and confirm the INIT | series prints, deltas are non-negative, and they sum to ~the existing "ready in {}ms" total. Use scripts/bench.sh + scripts/report_stats.py to capture deployed cold-start Init Duration deltas.

[datadog-official]

✨ Fix all issues with BitsAI

⚠️ Warnings

🚦 5 Pipeline jobs failed

DataDog/datadog-lambda-extension | integration-suite: [lmi]     

DataDog/datadog-lambda-extension | integration-suite: [on-demand]     

DataDog/datadog-lambda-extension | e2e-test-status (amd64)     

View all 5 failed jobs.

Useful? React with 👍 / 👎

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: d46a7ec | Docs | Datadog PR Page | Give us feedback!}