record_push_retry re-lowercases a static service name (heap allocation) on every retry attempt, on the provider-storm hot path

[erskingardner]

Summary

All three retry arms in retry.rs call metrics.record_push_retry(service_name.to_lowercase().as_str()). service_name is always a compile-time-constant &str ("APNs" or "FCM"), so to_lowercase() heap-allocates a fresh String on every retry attempt purely to lowercase a known-at-compile-time label. This runs precisely during provider 429/5xx storms and transport connect-error bursts — exactly the moments the retry path should stay allocation-light.

Location

• src/push/retry.rs:207 (auth-rejection retry)
• src/push/retry.rs:231 (transient 429/5xx retry)
• src/push/retry.rs:409 (transport connect-retry)

Callers always pass a static string: with_retry(..., "APNs", ...) (apns.rs:387), with_retry(..., "FCM", ...) (fcm.rs:560), with_transport_retry(..., "APNs", ...) (apns.rs:600), etc.

Details

Every other push metric already routes through platform_label(platform) (dispatcher/mod.rs:72), which returns the static lowercase "apns"/"fcm" with no allocation. Only the retry path re-derives the lowercase form per attempt via to_lowercase().

Recommendation

Pass the already-lowercase static label ("apns"/"fcm") into with_retry / with_transport_retry instead of the mixed-case "APNs"/"FCM", and drop the per-attempt to_lowercase(). (The mixed-case value is still used for the human-readable warn!(service = service_name, ...) log lines, so keep that or pass both.)

Why this is not a duplicate

#316 is about the metric label semantics (transport connect-retries and provider 429/5xx retries sharing one push_retry_total label). This is strictly the per-attempt allocation, an orthogonal concern.

Impact

Minor allocation churn confined to the retry path. LOW.