Details
-
Bug
-
Resolution: Fixed
-
High
-
4.5.0, 4.6.0, 4.7.0, 5.0, 5.1, 5.2-dev
-
None
-
ezp with DFS cluster using more than one web node
Description
the cache files are copied from the nfs server to the local node, and they get the timestamp of the local server.
assume this case:
- one installation using DFS, with webnodes A and B
- at time X, all caches are cleared on both servers
- a request someURL arrives to node A
- at time X+15min, the same request someURL goes to node B
at time X, caches are generated for nfs and node A. both servers get the copy of the cache file with a timestamp X
at X+15min, the server B makes the request, and the cache file is copied from nfs to the local var/ dir. the copy on server B will get a timestamp of X+15
now, assume that caches have a span of 1.00 hour. 1.05h later, 2 requests for the same someURL arrive at the same time, one lands on server A, another on server B.
the server A sees the cache as expired, will trigger a new cache generation. expected result.
the server B sees the cache still alive, and serves the old cache, until it reaches the 1.00 hour limit, which happens only at X+15min. so, there is a window of 15 minutes where the server B is serving the wrong cache
in more generic terms, in a multi node DFS cluster, the servers will retain the invalid caches for the interval between the cache generation and each server individual request.
Attachments
Issue Links
- relates to
-
EZP-20957 Misplaced semicolon causes flow to break
- Closed
- links to