- Bugfix: Ensure that if session.ping returns false we correctly identify fail the keepalive and connection
- Bugfix: Ensure that if the interval between keepalives being sent occurs faster than the prior keepalive's timeout that we do not overwrite the reference to the prior timeout. Prior implementation could have in theory prevented a valid keepalive timeout from clearing itself. This rewrite keeps every timeout as a local (vs a shared state per session). Even if the timeout outlives the lifetime of a session, we still guard against errors by checking that the parent interval is not false-y. I reckon this could result in a short-term memory leak per session which is bounded for a maximum of keepaliveTimeoutMs. On the other hand even with that potential for a short reference hold, this implementation proposed here is more correct I think. One alternative we could do is keep a list of pending timeouts.. which is complex for a rare situation that will self resolve anyhow when keepaliveTimeoutMs is reached.
- Bug Fix: keepalive intervals were being cleared with an incorrect clearTimeout before. Not sure if this was causing intervals leaks in some nodejs impls or not. (v20.13.1 seems to accept this mismatch without issue)
- Rename variables for clarity, to prevent future bugs like swapping clearInterval vs clearTimeout.
- Implementation is repeated in two places, per warning from grpc#2756 (comment)
- This commit supercedes the prior PR on a master branch which was out of date. grpc#2756

… first round of review from grpc#2760

…trace behavior

…gain

…epend on whether the session is destroyed

…ger. this is a necessary change to fit with having removed keepaliveDisabled boolean. manually inspected test logs for both server.ts and transport.ts to verify both types of keepalives are operating correctly.