Open robert-s-lee opened 6 years ago
Personal note: docker version is here: https://hub.docker.com/r/rijalati/oltpbench/
Currently testing and it looks like some benchmarks end prematurely.
Working
cockroach workload
)cockroach workload
)Not Working
The default runs for 1 minute. Maybe put -m 1000 for 1,000 seconds.
@robert-s-lee -- I was using a long-running -m
value, maybe its ignored for that one.
twitter
crashes after some time
13:28:02,432 (ThreadBench.java:283) INFO - Throughput: 1.6 txn/sec
13:28:07,433 (ThreadBench.java:283) INFO - Throughput: 1.4 txn/sec
java.lang.RuntimeException: Unexpected fatal, error in 'TwitterWorker<000>' when executing 'GetFollowers/03' [COCKROACHDB]
at com.oltpbenchmark.api.Worker.doWork(Worker.java:544)
at com.oltpbenchmark.api.Worker.run(Worker.java:296)
at java.lang.Thread.run(Thread.java:748)
Caused by: org.postgresql.util.PSQLException: This connection has been closed.
at org.postgresql.jdbc.PgConnection.checkClosed(PgConnection.java:783)
at org.postgresql.jdbc.PgConnection.rollback(PgConnection.java:791)
at com.oltpbenchmark.api.Worker.doWork(Worker.java:428)
... 2 more
Will have to debug those. The Twitter TPS seem very low. So 9 of the ones do work so far?
I wouldn't worry about the TPS -- I'm testing on two clusters to parallelize things. One cluster is running with the bare minimum resources (1 shared crdb + loadgen node), which is why tps is so slow. I'm just making sure each loadgen actually works.
Auctionmark is working now. So so far we have 10 generators that work. I'm almost done testing the rest
oltpbench's tpcc is too slow.
I'll use our cockroach workload
version instead
@nstewart just tried smallbank
You have to select a key when you use the template @robert-s-lee
Cloudformation doesnt have required fields so it fails with the message you saw
The other thing to watch out for is resource limits @robert-s-lee. If many SEs are running templates in the same region, you may bump into the 15 ec2 instance limit. If this becomes an issue we can see how to get these limits lifted.
If you hit an EC2 instance the autoscaling group will hang and eventually, the template will time out. if you hit a limit on VPCs/internet gateways/etc the template will fail immediately with an error.
https://docs.aws.amazon.com/general/latest/gr/aws_service_limits.html
made sure to fill in all of the items this time. nice sub millisecond response times on linkbench.
@robert-s-lee any update on seats
and twitter
issues?
oltpbenchmark has about 15 OLTP workloads that can be used for demo.
install java with `sudo apt-get -y update; sudo apt-get -y install openjdk-8-jdk maven'
install oltpbenchmark `git clone https://github.com/robert-s-lee/oltpbench.git --branch cockroachdb --single-branch oltpbenchmark; cd oltpbenchmark; ant'
run
cd oltpbenchmark;./oltpbehch.sh -w tpcc -t "1 2 4 8 16"
-w can have the following options
auctionmark linkbench seats tpcc tatp twitter voter epinions sibench smallbank ycsb wikipedia resourcestresser
-t is the concurrency. use
1
as default, but with an option to change-s is the size of the database. use
1
as the default, but with an option to changeThere are other options but we don't need to expose at this time.