Repository navigation
Conversation
|
this might require you to update your rust version to 1.95 if you're not already on that btw |
# Conflicts: # .github/workflows/rust.yml
# Conflicts: # Cargo.lock # odorobo/src/main.rs
…ecords on failed create, confirm VM deletion before removing durable state, agent no longer independently deletes manifests, add regression tests for pre-dispatch state
There was a problem hiding this comment.
protobuf-compiler was missing here; i don't remember why i changed this as part of this PR. but it needed to happen anyways
There was a problem hiding this comment.
re-enable rust-analyzer macro diagnostics
There was a problem hiding this comment.
install alpine's protoc package for cross-target container builds
There was a problem hiding this comment.
documentation on etcd key layout, versioned records, config fields, failure behavior, and why i decided to fail closed instead of using a non-durable local state on failure
There was a problem hiding this comment.
introduce versioned durable state abstraction with etcd, atomic create/delete of VM, placement, health checks, retries, optional TLS, and tests
There was a problem hiding this comment.
update cache-cleanup expectations for durable intent and supply an in-memory state store to scheduler tests
There was a problem hiding this comment.
recover a given node's assigned VM manifests from durable state at startup,restore resource accounting, reattach to or recreate VM actors, validate actor startup, and wait for delete confirmation before deleting durable records
There was a problem hiding this comment.
add CH heartbeat checks, stop actors after repeated VMM failures, avoid destroying an already destroyed instance, return deletion failures to callers
There was a problem hiding this comment.
detect an already running CH socket so a restarted agent can reattach instead of restarting a duplicate VMM. VMM deletion also requires confirmation before durable state gets removed.
There was a problem hiding this comment.
make VM deletion and heartbeat replies carry optional errors so callers can distinguish actor reachability from CH/VMM health
There was a problem hiding this comment.
This is missing code for durable VM shutdowns.
2 participants
#100