ConnectionWatchdog should be the last handler #335

Closed
mp911de opened this Issue Aug 17, 2016 · 0 comments

Projects

None yet

1 participant

@mp911de
Owner
mp911de commented Aug 17, 2016 edited

The reuse of handlers causes multiple calls to channelActive/channelInactive/and so on. In some cases, the reconnection is faster than channel deregistration (calls to channelUnregistered) which leads to the call sequence channelRegistered (new channel), channelUnregistered (old channel). Similar for channelActive/channelInactive. This is handled inside the CommandHandler but can still lead to issues. Moving the ConnectionWatchdog to the end reduces the probability of the calls coming in the wrong order. It also fixes the channelActive/channelInactive order in fast reconnect scenarios as the channelInactive call is done before notifying the ConnectionWatchdog.

@mp911de mp911de added this to the Lettuce 4.2.2 milestone Aug 17, 2016
@mp911de mp911de added a commit that referenced this issue Aug 17, 2016
@mp911de Add ConnectionWatchdog as last handler #335
The reuse of handlers causes multiple calls to channelActive/channelInactive/and so on. In some cases, the reconnection is faster than channel deregistration (calls to channelUnregistered) which leads to the call sequence channelRegistered (new channel), channelUnregistered (old channel). Similar for channelActive/channelInactive. This is handled inside the CommandHandler but can still lead to issues. Moving the ConnectionWatchdog to the end reduces the probability of the calls coming in the wrong order. It fixes also the channelActive/channelInactive order in fast reconnect scenarios as the channelInactive call is done before notifying the ConnectionWatchdog.
646115c
@mp911de mp911de added a commit that referenced this issue Aug 17, 2016
@mp911de Add ConnectionWatchdog as last handler #335
The reuse of handlers causes multiple calls to channelActive/channelInactive/and so on. In some cases, the reconnection is faster than channel deregistration (calls to channelUnregistered) which leads to the call sequence channelRegistered (new channel), channelUnregistered (old channel). Similar for channelActive/channelInactive. This is handled inside the CommandHandler but can still lead to issues. Moving the ConnectionWatchdog to the end reduces the probability of the calls coming in the wrong order. It fixes also the channelActive/channelInactive order in fast reconnect scenarios as the channelInactive call is done before notifying the ConnectionWatchdog.
ae70102
@mp911de mp911de closed this Aug 17, 2016
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment