Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
1.7.7
Description
Currently the standby server sends an un-persisted head record to clients. Under normal circumstances, the TarMK flush thread is able to persist it and its corresponding segment at a 5 seconds interval.
However, there are cases (uploading a very large blob > 10 GB) in which the flush thread writes the segment too late, and the 20s allowed by FileStoreUtil#readSegmentWithRetry are not enough. Therefore the server can't read the segment containing the head record and a timeout occurs on the client.