PROBLEM IN VERSION V4.6.0 RESOLVED BY V4.6.1
FIX FOR THE PROBLEM
The problem occured in some new code that we introduced in version V4.6.0 to optimize the operation of Heartbeat on the small minority of servers that have many (i. e. millions) of small files. In V4.6.1, this problem was fixed.
Under normal conditions when Heartbeat is replicating operations on a protected file, if Heartbeat can not access the protected file on the passive server then it will mark the file "out-of-sync." Heartbeat then resynchronizes the file between the active server and the passive server. We discovered a situation where the resynchronization operation can be queued for a long time, resulting in the potential for files to be "out-of-sync" for long periods. If the server roles are switched or if there is a failover during this period then the updates to those files queued for resynchronization could be lost. This issue only occurs with the version 4.6.0 of Heartbeat.
While the likelihood of the failure happening seems very small, the potential risk to the protected data is harder to estimate. In accordance with Neverfail’s policy of keeping our partners and customers informed, we made this information public.