2018-10-29 01:11:21 UTC - Rajan Dhabalia: hi @Matteo Merli / @Sijie Guo

Did you guys get chance to look into : 
<https://github.com/apache/pulsar/pull/2718>

It seems BK-4.7.2 is broken. We can't use wire-protocol with authentication 
enabled in bookie because of the issue mentioned into above PR.

Therefore, I tried to disable wire-protocol and I found the memory-leak in 
BK-4.7.2.
Because BK-client increments payload(data)'s ref-count but doesn't decrement 
when v2-protocol is disable : 
<https://github.com/apache/bookkeeper/blob/release-4.7.2/bookkeeper-server/src/main/java/org/apache/bookkeeper/proto/checksum/DigestManager.java#L129>
It's easily reproducible when we disable wire-protocol in broker's config.

So, it's hard to use BK-4.7.2 in both the scenarios.

Also, I saw memory-leak has been fixed in master with this commit: 
<https://github.com/apache/bookkeeper/commit/299fb58deed3dc284342716d52b7f918cc6cefc4#diff-7e13d6b039fa5c0f4739d817484fe858>
But I see a concern in that PR as well so, I have commented. So, can you also 
please take a look if I am missing anything here.
----
2018-10-29 01:19:10 UTC - Sijie Guo: @Rajan Dhabalia: #1585 
(<https://github.com/apache/bookkeeper/pull/1585>) is in 4.8.0 and onwards. It 
changes the ref-counting behavior (following netty’s pattern). so when we 
bumped to 4.8.0, we need to change pulsar’s code to retain before calling 
#asyncAddEntry. (I have one PR that changed that).

it might be easier to just fix the problem in 4.7.2 and we can cut a 4.7.3?
----
2018-10-29 01:20:34 UTC - Matteo Merli: yes, it would be good to have 4.7.3 
also for DNS caching issue and then cut 2.2.1
----
2018-10-29 01:21:48 UTC - Rajan Dhabalia: and is it possible to fix: 
<https://github.com/apache/pulsar/pull/2718>
in 4.7.3?
----
2018-10-29 01:24:01 UTC - Matteo Merli: @Rajan Dhabalia will look into it. I 
don’t remember the details at the moment
----
2018-10-29 01:24:10 UTC - Rajan Dhabalia: sure..thanks
----

Reply via email to