Subscriber unsubscribes after subscribing on reconnect

[shellicar]

Issue

Subscriber very occasionally unsubscribes after subscribing after the subscriber connection is restored from a hard disconnect (port being closed).
This seems to be related to issue: #273
But I can still reproduce this.

Version

Reproduced with the following versions.
1.2.6
2.0.519
2.0.588

Behaviour

using redis-monitor on the server, the following is observed.

1554338345.603361 [0 127.0.0.1:10441] "SUBSCRIBE" "__Booksleeve_MasterChanged"
1554338345.620982 [0 127.0.0.1:10441] "SUBSCRIBE" "Channel1"
1554338345.621008 [0 127.0.0.1:10441] "SUBSCRIBE" "Channel2"
1554338345.629399 [0 127.0.0.1:10441] "UNSUBSCRIBE" "Channel1"
1554338345.629421 [0 127.0.0.1:10441] "UNSUBSCRIBE" "Channel2"

Expected/Normal Behaviour

1554338954.604720 [0 127.0.0.1:11370] "SUBSCRIBE" "__Booksleeve_MasterChanged"
1554338954.609184 [0 127.0.0.1:11370] "UNSUBSCRIBE" "Channel1"
1554338954.609228 [0 127.0.0.1:11370] "UNSUBSCRIBE" "Channel2"
1554338954.609439 [0 127.0.0.1:11370] "SUBSCRIBE" "Channel1"
1554338954.609456 [0 127.0.0.1:11370] "SUBSCRIBE" "Channel2"

OR

1554338956.418663 [0 127.0.0.1:11372] "SUBSCRIBE" "__Booksleeve_MasterChanged"
1554338956.419116 [0 127.0.0.1:11372] "SUBSCRIBE" "Channel1"
1554338956.419133 [0 127.0.0.1:11372] "SUBSCRIBE" "Channel2"

[shellicar]

Any information on this?
I should add that the internal subscribers dictionary still thinks its subscribed.

This makes the library unusable for us as the subscriptions are unreliable.

[mgravell]

I haven't had chance to look (all support here comes from our personal time, etc), but I agree it looks incorrect. My suspicion is that it *thinks* it is unsubscribing from the old primary - most likely what I need to do here is track the endpoint currently subscribed, and make sure we include that in all routing decisions.

…

On Fri, 19 Apr 2019, 04:52 Stephen Hellicar, ***@***.***> wrote: Any information on this? I should add that the internal subscribers dictionary still thinks its subscribed. This makes the library unusable for us as the subscriptions are unreliable. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1110 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAEHMBCEZYCQSP5UHLG7ZLPRE6XBANCNFSM4HDO7VPA> .

[shellicar]

No problem, I appreciate any work done in peoples personal time.

I would have thought it was an error with how the events were invoked/queued, as they are correct, just occur out of order.

[NickCraver]

@mgravell interesting race to note as we dig on this area - tagging for #1912

[NickCraver]

This is drastically improved in #1947 and will be in the 2.5 release - thank you for the detail!