Replay Sessions

A replay session is an interactive reconstruction of an incident timeline.

Replay helps engineers answer:

  • What happened first?
  • Which delay pushed the request over its timeout?
  • What if a timeout was longer?
  • What if network latency was higher?
  • What if a specific failure happened earlier?

Timeline replay

The baseline replay mode plays recorded events in their original order and timing.

event 1 at +0ms
event 2 at +2ms
event 3 at +842ms
event 4 at +850ms

You can pause, seek, slow down, or speed up the incident.

Timing injection

kprobe can run a what-if replay with alternate timing:

  • add network latency
  • multiply storage latency
  • override timeout budgets
  • inject a failure at a selected event

Full syscall replay

For supported workloads, kprobe can use recorded syscall behavior to reproduce a process-level incident in a controlled sandbox. This mode is designed for deep debugging and validation of timing-sensitive fixes.

When to use replay

Use replay after a causal graph identifies a plausible cause. The graph tells you where to look. Replay helps you test whether the timeline supports the hypothesis.