Re:no response for get children

[happen <37...@qq.com>]

Hi,

It turns out to be ZOOKEEPER-3072 still in zk 3.5.6. (https://issues.apache.org/jira/browse/ZOOKEEPER-3072)

Last month, we set a high globalOutstandingLimit value  as a temporary measure to avoid this problem until next version zk published to our production environment.


------------------ Original ------------------
From:                                                                                                                        "happen"                                                                                    <37...@qq.com>;
Date: Sun, Aug 16, 2020 10:24 PM
To: "dev"<de...@zookeeper.apache.org>;
Subject: Re: no response for get children 

Hi,    
        
I found a strange thing in zk 3.5.6 that sometimes observers ignored one get children req (local session),  thougth other watch events could still be sent to the involved c++client. 

Then the involved clientwaited the get children reponse for totally session timeout time and no ping could be sent to observer until the session expired. The client said "invalid zhandle state" after session expired.  
         
The issue's frequency situation:               
 A)  1 to 5 get children requests are ignored per minutewhen:6000 client is connected to one observer, zk_packets_sent 2M/min, zk_packets_received 1M/min;     
 B) and less get children requests are ignored when lower zk_packets_sent and zk_packets_received.  

I caught packets for the journey, no one reponses for the ignored get children req when it occurred.Why  sometimes get children req.s were ignored ?Is it possible a bug in zk3.5.6 ?