Fix for stale session issue

[colinmcintosh]

Added logic to reset the client SessionID if the client disconnects and does not reconnect before the session timeout time elapses. make test passes for me with these changes.

Fixes #36.

[codecov]

Codecov Report

Merging #37 (127b9eb) into master (50daf81) will decrease coverage by 0.85%.
The diff coverage is 83.33%.

@@            Coverage Diff             @@
##           master      #37      +/-   ##
==========================================
- Coverage   77.58%   76.73%   -0.86%     
==========================================
  Files           7        7              
  Lines        1316     1199     -117     
==========================================
- Hits         1021      920     -101     
+ Misses        205      190      -15     
+ Partials       90       89       -1     

Impacted Files	Coverage Δ
conn.go	74.30% <83.33%> (-0.31%)	⬇️
constants.go	75.00% <0.00%> (-6.25%)	⬇️
lock.go	63.38% <0.00%> (-4.12%)	⬇️
dnshostprovider.go	72.72% <0.00%> (-1.64%)	⬇️
structs.go	80.75% <0.00%> (-1.28%)	⬇️
flw.go	82.78% <0.00%> (-0.97%)	⬇️
util.go	93.61% <0.00%> (+0.75%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 50daf81...127b9eb. Read the comment docs.

[yarikk]

Hi, @colinmcintosh, thanks for your submission. Do you think you could provide a test which exercises those code paths?

[pmazzini]

I am trying to understand what is this change fixing.

Doesn't the session get reset in here for you https://github.com/go-zookeeper/zk/blob/master/conn.go#L673?

[zmstone]

@pmazzini
judging by the EOF error (full line: authentication failed: EOF)
it's likely the code is returned from somewhere before L673
maybe here: https://github.com/go-zookeeper/zk/blob/master/conn.go#L663-L666

[nemith]

Any way we can get a unit test on this?

[killerwhile]

If this can help someone, I got this authentication failed: EOF in a cluster running v3.5.8 and apparently in a state like described in https://issues.apache.org/jira/browse/ZOOKEEPER-3829 and https://issues.apache.org/jira/browse/ZOOKEEPER-3830. Restarting the nodes did help a lot.

[colinmcintosh]

Hi! I pushed a couple updates to restrict this session reset further and moved the disconnect time to fix an issue with tight loops not being caught by the timer.

I did attempt to write a test for this that started, stopped, and started the test ZK server to reset the session but couldn't figure out how to make it work with the test server. I'm able to easily reproduce this by running Zookeeper in a Docker container and restarting the container (docker kill zookeeper && docker run --rm --name=zookeeper -d -p 2181:2181 zookeeper). I'd appreciate any help writing tests to copy that behavior; it would be great to get this PR merged in since it's been hanging around for a while.

[jpfourny]

I will take a stab at a unit test to produce the issue when I find some magic 🦄 time. 🤣 No promises.

[joeedt]

i think this should place after line 489;

when zookeeper cluster re-elect, reconnect always success, but auth failed:(EOF). line 493 will rfresh disconnectTime every loop; line 442 always be False

[LittleCuteBug]

Hi @colinmcintosh, are you still working on this pull request. I'm facing similar issue and hoping this pull request will be done.

[covy-blue]

Hi, I am still experiencing this issue.