Reconnect behavior

When the SSH connection drops mid-session, the plugin retries up to 5 times with ×1.5 exponential backoff (1 s → 1.5 s → 2.25 s → … capped at 30 s) before surfacing an error.

Default policy

• Trigger: any RPC call that fails with a transport-layer error (broken pipe, EOF, connection reset).
• Retries: 5 (configurable: Settings → Advanced → Reconnect attempts).
• Backoff: starts at 1 s, multiplier ×1.5, ±20% jitter, capped at 30 s per attempt — so nominal: 1 s, 1.5 s, 2.25 s, 3.4 s, 5.1 s, …
• What’s reused: the daemon. The plugin re-opens the SSH session and the local TCP forward, then re-auths and re-subscribes to any active watchers.

What you’ll see

A subtle banner appears at the bottom of the workspace:

Reconnecting to "My Pi" (attempt 2/5)…

• Within 5–10 seconds (typical home WiFi blip): banner disappears, life continues.
• After all retries exhausted: banner becomes red — “Disconnected from My Pi”, with a “Retry” button. The shadow vault stays open so you don’t lose unsaved work; writes queue locally and flush on next connect.

What about the daemon

The daemon stays running. If the SSH session is the only thing that died (kernel didn’t kill the daemon), reconnect is fast — same daemon, same socket, same token. If the OS rebooted between drops, the daemon is gone too; the plugin re-deploys it as part of the reconnect attempt (~5 seconds).

Disabling auto-reconnect

Set Reconnect attempts to 0 in advanced settings. Drops will surface immediately as errors. Useful if you’re debugging a flapping link and want raw signal.

Mobile / sleep behavior

When your laptop sleeps, the SSH session stays “open” from your end’s perspective — the kernel doesn’t actively close it. On wake, the first RPC will hang for whatever your TCP keepalive is configured to, then fail with a broken pipe. The reconnect kicks in then.

For long sleeps (overnight), this often means a 30–60 second wait on first interaction after wake. To shorten: enable TCP keepalive on the remote sshd (ClientAliveInterval 60 in sshd_config).

Watcher resubscription

After reconnect, the plugin re-issues any active fs.watch calls. Events that fired during the disconnect window are NOT replayed — the plugin instead does a one-shot fs.walk to compare and surface anything that changed. This catches “I edited a file on the remote while my laptop was asleep”.

For very long disconnects (hours), the comparison is bounded by maxEntries (default 50,000). If you exceed it, the plugin shows a notice and offers a manual “rescan vault” action.

Next: Daemon panel.