7e72ceffd1
Multiple spots were not using DB transactions when processing the terminal
state transitions (error, abort, finish, timeout). The pattern looked like
this:
node_info.fsm_event(istate.Events.error)
# more code
node_info.finished(error='Oops!')
which led to brief periodes of state inconsistency of NodeInfo records in
the DB.
This patch refactors the NodeInfo.finished() method to require a terminal state
transition to perform as part of the NodeInfo state update:
NodeInfo().finished(istate.Events.finish)
NodeInfo().finished(istate.Events.abort, 'Canceled by operator')
This patch also introduces a new state: aborting to allow the inspector to
try call power-off the node before marking the introspection aborted.
There's a new DB migration since the new state implies a schema change too
(Enum).
Closes-Bug: #1721233
Closes-Bug: #1721230
Closes-Bug: #1723384
Change-Id: I0bb051d1956a996ed006d55a5ca2d670d9455047
13 lines
492 B
YAML
13 lines
492 B
YAML
---
|
|
upgrade:
|
|
- |
|
|
A new state ``aborting`` was introduced to distinguish between the node
|
|
introspection abort precondition (being able to perform the state
|
|
transition from the ``waiting`` state) from the activities necessary to
|
|
abort an ongoing node introspection (power-off, set finished timestamp
|
|
etc.)
|
|
fixes:
|
|
- |
|
|
The ``node_info.finished(<transition>, error=<error>)`` now updates node
|
|
state together with other status attributes in a single DB transaction.
|