Comment 2 for bug 1168399

Revision history for this message
Brant Knudson (blk-u) wrote :

Interestingly, I was working on a system where this fix would have been useful (as a workaround) today.

The test was for stability/scalability, so it was running for several hours booting instances. Keystone's token table got to about 1 million rows, mostly because nova using quantum creates a new token for every quantum op as far as I can tell.

Eventually, getting the revoked tokens list takes ~9 seconds, and then nova fails because quantum's getting token times out (so they blame Keystone of course).

Creating the index made it so that the query to get the revoked tokens list now takes < 1 second again.

Of course, the real fix is that nova/quantum should re-use its token rather than getting a new one every time.