[ernado]

Notes from running gotd in a long-running, multi-account archiver. Each one explains behaviour that already exists.

Update recovery (helpers/updates-recovery.md):

• Your own sends. The results of sendMessage, editMessage and similar calls reach only the calling session. Without UpdateHook your handler sees them late. Even with it, updateShortSentMessage becomes MessageEmpty. The page includes a middleware that rebuilds the message from the request.
• "Failed to load channel state". Explains what the warning means (a stored pts with no access hash) and how to fix it (the hasher, or TrackChannel).
• Channels that aren't pushed. They're polled every 15 minutes (channel-idle-timeout), so they arrive up to 15 minutes late. Covers how the official clients handle this and how to spot it in the metrics.

Long-running clients (advanced/long-running.md), lost authorization in practice:

• a terminated session shows up as AUTH_KEY_UNREGISTERED;
• the loss is detected on the next request that needs authorization;
• with a shared errgroup, one account's AuthLostError stops every account;
• how to test it safely on a spare account.

Observability (advanced/observability.md):

• several clients on one MeterProvider, separated with a scope attribute per client (otlp.promote_scope_metadata for Prometheus);
• why the state counter can't tell you whether a client is connected right now.

Facts checked against td main: the 15-minute idle timer, which resets on each channel update; loadChannels; convertShortSentMessage; the UpdateShortSentMessage fields; MiddlewareFunc. yarn build passes.

Banned, deleted and frozen accounts

Documents gotd/td#1934 (AuthLostKindOf, AUTH_KEY_DROP_*, OnFrozen, FreezeState, gotd.client.account.restricted). Merge after #1934 is merged.