[mmoulikk]

Issue

While a new SF Compute node is starting, SF Compute already lists it as running, but GET /integrations/brev/v1/instances/{id}/ssh returns 404 ("SSH details not currently available") for a few seconds. ListInstances left the node out of the list, and dev-plane reads a missing node as terminated. The environment then moved to instance_lost and could end up in terminated: setup never ran, Delete was hidden in the console, and the VM kept running and billing (BREV-13508).

Root cause

getSSHInfo returned every SSH lookup error to its caller. ListInstances skipped any node whose lookup failed, and GetInstance returned the error. A 404 right after creation means the SSH details aren't ready yet, not that the node is gone.

Fix

• getSSHInfo: a 404 within 45 minutes of creation (sshDetailsPendingWindow) means the node is still starting. The node stays listed as pending, GetInstance no longer errors, and it logs sfcv2: SSH details not available yet, reporting instance as pending.
• sfcInstanceToBrevInstance: a running node without an SSH hostname is reported as pending, since Brev can't reach it yet.
• After 45 minutes, and for any other error (429, 5xx, connection), nothing changes and the node is skipped as before. 45 minutes is longer than the build's SSH wait (up to 3 × 10 min), so a node that never gets SSH details still fails its build.
• pending doesn't change an environment's state in dev-plane, and time in pending isn't billed.

Test

• Unit tests in instance_test.go, against a fake SF Compute API:
  • TestIsSSHDetailsPending: a 404 inside the window counts as pending; a 404 after the window, a 500, or a connection error doesn't.
  • TestListInstancesKeepsNewInstanceWithoutSSHDetailsAsPending: a new node with an SSH 404 is listed as pending, a ready node is running with its SSH address, and an old node with an SSH 404 is still left out.
  • TestGetInstanceReportsNewInstanceWithoutSSHDetailsAsPending: GetInstance returns pending instead of an error.
• CI passes.

Staging (dev-plane v0-brev-13508-0, Nov 13)

• Created 11 SF Compute environments. 8 nodes hit the SSH 404 and were kept as pending instead of dropped (logs).
• No node was dropped after the deploy. The only "skipping instance due to SSH error" came from the previous build, at 11:05 UTC (log).
• No environment went to instance_lost.
• dau23x56n (logs) and coyxnym6e (logs) hit the 404, stayed in building, and reached running after one build retry while SSH came up. The other 9 were deleted on purpose mid-build, as part of the test.