Message accounting end to end
reliability
Outbound platform across Email, SMS, App, and WhatsApp. Source and target counts never matched.
False drops made every investigation a fight about the logs. We split metrics by stage, killed the fake failures,
and got to 100% accounting source to target.
note →
· timeline →
Cost, quality, and a retry storm
operations
Cut AWS cost about 21% by removing network and disk bottlenecks. Lifted SonarQube quality metrics 78% across services.
Redesigned WhatsApp as a bot-of-bots. Separate story: one slow Aerospike shard turned into an outage when retries lined up.
Fix was skip bad keys, then backoff with jitter.
retry storm note →
· timeline →