William Bardwell created TS-1703:
------------------------------------
Summary: simultaneous cache entry refresh has performance issues
Key: TS-1703
URL: https://issues.apache.org/jira/browse/TS-1703
Project: Traffic Server
Issue Type: Improvement
Components: Cache, HTTP
Reporter: William Bardwell
If you use enable_read_while_writer off, any requests after the first that
occur while a cache entry needs to be refreshed end up going to the origin
(internally the second request treats things as a cache miss.)
If you use enable_read_while_writer on, then a lot of requests are answered
from a single refresh, but occasionally some will have trouble getting some
sort of write lock and will still fail, and will go to the origin when they
shouldn't. (And they will trigger TS-1702 until that is fixed.)
I would argue that for both lock failure cases, instead of just ignoring the
cache entry, it should be used in a read-only mode and should still do a
refresh, and just use the entry if a 304 is received or use the new contents
(but don't save them, since it doesn't have a write lock) if it does not get a
304.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
