On 11/13/2014 02:44 PM, Andres Freund wrote:
> Hi Steve,
> If it still happens, could you send me instructions of how to reproduce
> the problem after cloning the necessary source repositories? It's quite
> hard to validate a possible fix otherwise.
1. Install PG 9.4
2. Perform an initdb max_connections = 200 wal_level=logical max_walsenders=50 wal_keep_segments = 100
wal_sender_timeout= 600s max_replication_slots = 120
Create a user slon with superuser
create a user test
(set passwords for them you'll use them later)
Configure pg_hba.conf to allow slon to connect as a replication user
3. Download slony from github (https://github.com/ssinger/slony1-engine.git) checkout the branch
logical_remote_helper ./configure --with-pgconfigdir=/path_to_your_pgcbindir
make
make install
4. Download clustertest & compile clustertest from
(https://github.com/clustertest/clustertest-framework).
5. In the slony source tree cd to the clustertest directory
6. cp conf/disorder.properties.sample to conf/disorder.properties
Edit the file to have the correct paths, ports, users, passwords.
7 cp conf/java.properties.sample to conf/java.properties edit the file
to point at the JAR you downloaded earlier.
I run with all 5 databases on the same PG cluster. Performance will be
much better with 5 different clusters, but I recommend trying to emulate
my configuration as much as possible to replicate this
To run the tests then do
./run_all_disorder_tests.sh
At the moment (commit df5acfd1a3) is configured to just run the Failover
node test cases where I am seeing this not the entire suite.
It typically takes between 30 minutes to an hour to run though.
I installed things following the above steps on a different system than
my usual development laptop and I have been unable to reproduce the
error so for (on that system). But I am still able to reproduce it on
occasion on my normal development laptop.
> Greetings,
>
> Andres Freund
>
