CaptureTool (built next to the servers) replays packet bundles and compares the answers:
- replay: per bundle a fresh sandbox folder with copied server binaries, rewritten settings
(replay_sandbox=1, a new SQLite file inside the folder, ports from --port on, no dashboard),
read back before anything starts; sandbox-setup runs inside it to apply migrations and make
the replay account and the bundle's characters (setup mode); master is started, the stack is
stopped as one process group and the folder deleted unless kept
- with replay_sandbox=1 every server refuses a database that isn't SQLite, isn't inside its
own folder, or is replay_live_sqlite_path (Database::Connect); replay-target against a
running server needs --i-know-this-is-not-a-sandbox
- the fake client splits the recording into connections, logs in and picks the character
itself when the recording doesn't, fills in the target's account, session key and IDs,
learns server-made object IDs from replica constructions by LOT, follows the recorded
timing and waits for the answers a client waits for; the diff pairs answers by name (and
constructions by LOT) and ignores fields that differ between runs
- import-live converts the 2014 live captures (folders of *_traffic.zip; pcaps and encrypted
captures are left alone) into bundles, with secrets removed and CREATE_CHARACTER as setup
- anonymise makes local fixtures; docs/CaptureReplay.md describes capture, the bundle format,
portability rules, the sandbox and the replay
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
