java client doesn't work after network split

[umatomba]

Hi Robert,

trying to test the 5 node cluster again....

Space ycsb configuration:

space usertable
key k
attributes field0, field1, field2, field3, field4,
           field5, field6, field7, field8, field9
tolerate 2 failure

Workload used:

recordcount=1000000
operationcount=500000
workload=com.yahoo.ycsb.workloads.CoreWorkload
readallfields=true
readproportion=0.1
updateproportion=0.9
scanproportion=0
insertproportion=0
requestdistribution=zipfian
threadcount=1

After the cluster setup the config looks like:

server 1998750160347642936 192.168.100.3:2012 AVAILABLE
server 3667604389049738774 192.168.100.5:2012 AVAILABLE
server 6903549772667245983 192.168.100.2:2012 AVAILABLE
server 12441010121126734978 192.168.100.1:2012 AVAILABLE
server 12691374307851476699 192.168.100.4:2012 AVAILABLE

Now we can see all 1,000,000 elements in usertable, fine:

>>> c = hyperdex.client.Client('192.168.100.1', 1982)
>>> c.count('usertable', {'k': hyperdex.client.Regex('.*')})
1000000L

Lets start the test:

java -Djava.library.path=/usr/local/lib/ com.yahoo.ycsb.Client -t -db org.hyperdex.ycsb.HyperDex -P /hyperdex
_build/ycsb-0.1.4/workloads/workloada -p "hyperdex.host=192.168.100.1" -p "hyperdex.port=1982" -s

Making the network split for 10 sec every 30 sec on nodes 4,5 we see this log:

 10 sec: 23924 operations; 2388.82 current ops/sec; [UPDATE AverageLatency(us)=440.79] [READ AverageLatency(us)=107.13]
 20 sec: 49520 operations; 2559.34 current ops/sec; [UPDATE AverageLatency(us)=415.98] [READ AverageLatency(us)=97]
 30 sec: 72777 operations; 2325.7 current ops/sec; [UPDATE AverageLatency(us)=466.64] [READ AverageLatency(us)=91.18]
 40 sec: 93333 operations; 2055.39 current ops/sec; [UPDATE AverageLatency(us)=524.7] [READ AverageLatency(us)=96.18]
 50 sec: 130622 operations; 3728.9 current ops/sec; [UPDATE AverageLatency(us)=275.01] [READ AverageLatency(us)=72.35]
 60 sec: 146083 operations; 1546.1 current ops/sec; [UPDATE AverageLatency(us)=595.3] [READ AverageLatency(us)=72.74]
 70 sec: 146148 operations; 6.5 current ops/sec; [UPDATE AverageLatency(us)=193092.31] [READ AverageLatency(us)=19839.25]
 80 sec: 146286 operations; 13.8 current ops/sec; [UPDATE AverageLatency(us)=78855.65] [READ AverageLatency(us)=757.45]
 90 sec: 146352 operations; 6.6 current ops/sec; [UPDATE AverageLatency(us)=14269.27] [READ AverageLatency(us)=119.57]
 100 sec: 146437 operations; 8.5 current ops/sec; [UPDATE AverageLatency(us)=239552.85] [READ AverageLatency(us)=3672.67]
 110 sec: 146688 operations; 25.1 current ops/sec; [UPDATE AverageLatency(us)=41942.16] [READ AverageLatency(us)=2646.36]
 120 sec: 146783 operations; 9.5 current ops/sec; [UPDATE AverageLatency(us)=79886.51] [READ AverageLatency(us)=995.67]
 130 sec: 146783 operations; 0 current ops/sec;
 140 sec: 146783 operations; 0 current ops/sec;
 150 sec: 146783 operations; 0 current ops/sec;
 160 sec: 146783 operations; 0 current ops/sec;
 170 sec: 146783 operations; 0 current ops/sec;
 180 sec: 146783 operations; 0 current ops/sec;
 190 sec: 146783 operations; 0 current ops/sec;
 200 sec: 146783 operations; 0 current ops/sec;
 210 sec: 146783 operations; 0 current ops/sec;
 220 sec: 146783 operations; 0 current ops/sec;
 230 sec: 146783 operations; 0 current ops/sec;
 240 sec: 146783 operations; 0 current ops/sec;
 250 sec: 146783 operations; 0 current ops/sec;
 260 sec: 146783 operations; 0 current ops/sec;
 270 sec: 146783 operations; 0 current ops/sec;
 280 sec: 146783 operations; 0 current ops/sec;
 290 sec: 146783 operations; 0 current ops/sec;
 300 sec: 146783 operations; 0 current ops/sec;
 310 sec: 146783 operations; 0 current ops/sec;

And now, after the cluster restart, java client VM restart (it's one separate VM) we can't make the ycsb test with this cluster:

root@dbtest0:/hyperdex_build/tests/5nodes/teste# java -Djava.library.path=/usr/local/lib/ com.yahoo.ycsb.Client -t -db org.hyperdex.ycsb.HyperDex -P /hyperdex
_build/ycsb-0.1.4/workloads/workloada -p "hyperdex.host=192.168.100.5" -p "hyperdex.port=1982" -s > out5
Loading workload...
Starting test.
host: 192.168.100.1
port: 1982
 0 sec: 0 operations;
m_client: org.hyperdex.client.Client@6342f024

 10 sec: 204 operations; 20.37 current ops/sec; [UPDATE AverageLatency(us)=536.47] [READ AverageLatency(us)=131.33]
 20 sec: 204 operations; 0 current ops/sec;
 30 sec: 204 operations; 0 current ops/sec;
 40 sec: 204 operations; 0 current ops/sec;
 50 sec: 204 operations; 0 current ops/sec;
 60 sec: 204 operations; 0 current ops/sec;
 70 sec: 204 operations; 0 current ops/sec;

It looks as java-client problem, because the python client works fine...

Any ideas?

Thank you,
Vladimir

[hwinkel]

thats really annoying, especially as we have not much interest in the Java Client apart of the YCSB Tests. Is there any suggestion about the most robust Clients to use?