Hello,
I will comment the request cache section.
If I shutdown the backend is still active, but it has no more active
connections. After a while, the activity is back. While the activity is
stopping I see that logs are still registered. But I notice that
activity can be stopped on all my controller nodes.
+--------------------------+-------------------------------------+
| Backend Name | xxx-1 |
| Driver | org.postgresql.Driver |
| URL | jdbc:postgresql://IP/xxx|
| Active transactions | 0 |
| Pending Requests | 0 |
| Read Enabled | true |
| Write Enabled | true |
| Is Initialized | true |
| Static Schema | false |
| Connection Managers | 1 |
| Total Active Connections | 5 |
| Persistent Connections | 0 |
| Total Requests | 11254 |
| Total Transactions | 7301 |
| Last known checkpoint | <unknown> |
+--------------------------+-------------------------------------+
withouth request cache section ... it's much better.
Regards,
Arnaud.
Emmanuel Cecchet a écrit :
Hi Arnaud,
I do not recommend the use of the ResultCache in general cases. Its
implementation is not production quality.
<ResultCache granularity="table" />
I will need some clarifications on the following.
Normal usage of the cluster is good. My databasses are consistent.
You mean that all databases see the updates when they are all enabled,
correct?
My test scenario : I shutdown my db #3. Controller #3 notice it.
Well, if you disable backend 3 on controller 3, all controllers should
create a checkpoint (you can see the list of checkpoints in expert
mode I think). If not, it means that you have a group communication
problem. Or did you try to shutdown the virtual database on controller
3? The exact commands used would help diagnose the problem.
System is still running. I restart my database. I try to re enable
my backend (after a dump restore) ... no success. then I must restore
logs from another node. Then it's the end. The node suspend my vdb
activity ... no more after 20 mins.
This sounds like a group communication problem. The group
communication uses generally (if configured with UDP) both multicast
to send message and unicast for acknowledgments and other signaling
messages. There might be a problem there.
Keep us posted with your progress,
Emmanuel
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
