Ensure idl.run() called on TRY_AGAIN

If the DB connection drops after Connection.run() calls idl.run()
but before Transaction.do_commit(), then it is possible for
idl.commit() to return TRY_AGAIN due to the failure of session.send().
This prevents commit_block() from running idl.run() which would
actually trigger the DB reconnect, causing each successive call to
commit_block() to fail in the same way until we stop trying again
due to timeout and Connection.run() calls idl.run() again.

For now, we can just ensure that idl.run() is called when we get a
TRY_AGAIN. It might be worth it to refactor this a bit to not even
use commit_block().

Change-Id: I77402bde3fc4d45c770b9f48882870cfc91d719d
Closes-Bug: #1741889
(cherry picked from commit 1810faecc9)
This commit is contained in:
Terry Wilson 2018-01-24 03:41:13 +00:00
parent 742754bce3
commit e5ded80fe5
1 changed files with 7 additions and 0 deletions

View File

@ -91,6 +91,13 @@ class Transaction(api.Transaction):
status = txn.commit_block()
if status == txn.TRY_AGAIN:
LOG.debug("OVSDB transaction returned TRY_AGAIN, retrying")
# In the case that there is a reconnection after
# Connection.run() calls self.idl.run() but before do_commit()
# is called, commit_block() can loop w/o calling idl.run()
# which does the reconnect logic. It will then always return
# TRY_AGAIN until we time out and Connection.run() calls
# idl.run() again. So, call idl.run() here just in case.
self.api.idl.run()
continue
elif status == txn.ERROR:
msg = "OVSDB Error: %s" % txn.get_error()