Consumer Messaging Platform — WhatsApp-Class System  ·  View 23 of 23

Observability and Service Levels

Six signal classes across five pipeline stages, each ending in a defined action — the matrix form makes a missing cell visible.

Editable source SVG draw.io All views
Emit
Emit
Collect
Collect
Store
Store
Evaluate
Evaluate
Act
Act
Availability
Availability
Gateway and service health
Gateway and service health
OpenTelemetry agents
OpenTelemetry agents
Metrics TSDB
Metrics TSDB
99.99% messaging path
99.99% messaging path
Page on-call, shed load
Page on-call, shed load
Latency
Latency
Send to deliver spans
Send to deliver spans
Trace sampling 1 in 1000
Trace sampling 1 in 1000
Trace store
Trace store
p99 under 500 ms in region
p99 under 500 ms in region
Halt rollout, scale fanout
Halt rollout, scale fanout
Delivery integrity
Delivery integrity
Ack and receipt counters
Ack and receipt counters
Event bus
Event bus
Delivery ledger
Delivery ledger
Undelivered over 30 min
Undelivered over 30 min
Replay from device inbox
Replay from device inbox
Abuse & safety
Abuse & safety
Reports, spam and rate signals
Reports, spam and rate signals
Safety pipeline
Safety pipeline
Abuse record store
Abuse record store
Ban score threshold
Ban score threshold
Throttle, then ban
Throttle, then ban
Capacity & cost
Capacity & cost
Socket, fanout and egress counts
Socket, fanout and egress counts
Metrics agents
Metrics agents
Capacity warehouse
Capacity warehouse
Headroom under 30%
Headroom under 30%
Scale gateway pools
Scale gateway pools
Client health
Client health
Crash, ANR and battery reports
Crash, ANR and battery reports
Crash pipeline
Crash pipeline
Crash store
Crash store
Crash-free 99.9% of sessions
Crash-free 99.9% of sessions
Staged client rollback
Staged client rollback
Observability and Service Levels
Observability and Service Levels
Signals are never derived from message content; every row is metadata or client telemetry.
Signals are never derived from message content; every row is metadata or client telemetry.
v 1.0 · owner Platform Architecture · date 2026-08
v 1.0 · owner Platform Architecture · date 2026-08Text is not SVG - cannot display

Service levels

  • Messaging path availability 99.99%, measured at the gateway, not at the load balancer
  • Send-to-deliver p99 under 500 ms in region; cross-region measured separately
  • Crash-free sessions 99.9%, which gates client rollout progression

Delivery integrity

  • Acknowledgement and receipt counters are reconciled continuously against the delivery ledger
  • Anything undelivered beyond 30 minutes with an online device raises an alarm
  • The remedy is a replay from the device inbox, which is always safe

Privacy constraint

  • No signal is derived from message content, because none can be
  • Every row is metadata, counters or client telemetry
  • This constrains debugging: reproduction depends on client-side diagnostics the user opts into
Solution Architecture v1.0 · Platform Architecture · 2026-0823-observability-slo