| *** J1m_ has quit IRC | 01:02 | |
| *** fdrake has quit IRC | 01:23 | |
| *** jaroel|4D has quit IRC | 09:37 | |
| *** jaroel|4D has joined #zodb | 10:17 | |
| *** J1m_ has joined #zodb | 14:14 | |
| *** J1m_ has quit IRC | 14:15 | |
| *** srichter_ has quit IRC | 14:22 | |
| *** fdrake has joined #zodb | 14:50 | |
| *** srichter_ has joined #zodb | 15:24 | |
| *** J1m_ has joined #zodb | 15:40 | |
| *** povbot has joined #zodb | 17:00 | |
| *** srichter_ is now known as srichter | 18:16 | |
| *** J1m_ has quit IRC | 18:42 | |
| *** J1m_ has joined #zodb | 19:00 | |
| *** srichter has quit IRC | 20:17 | |
| *** jaroel|4D has quit IRC | 20:32 | |
| *** jaroel|4D has joined #zodb | 20:32 | |
| *** jaroel|4D has quit IRC | 20:37 | |
| *** srichter has joined #zodb | 21:09 | |
| *** goschtl has joined #zodb | 22:24 | |
| goschtl | hi, does it make sense if i add more zeo-clients to enhance write performance for zodb? | 22:25 |
|---|---|---|
| J1m_ | not much | 22:28 |
| J1m_ | depending | 22:28 |
| J1m_ | a few is better than one | 22:28 |
| J1m_ | many isn't likely to be better than a few. | 22:28 |
| goschtl | ok i have now three | 22:29 |
| goschtl | let's see how this will work | 22:29 |
| J1m_ | 3 is probably better than one. | 22:29 |
| J1m_ | 6 could be better than 3. | 22:29 |
| goschtl | i will play with it | 22:29 |
| J1m_ | The first phase of two-phase commit can happen in parallel. | 22:30 |
| J1m_ | (It should be possible to allow some more parallelism in the second phase too, but I haven't gotten to that yet.) | 22:30 |
| goschtl | 2013-12-03 13:51:16,435 WARNING [ZopePublication] Competing writes/reads | 22:31 |
| goschtl | i have received a lot of these today | 22:31 |
| goschtl | it turns out the the intid utility was the bottleneck | 22:31 |
| J1m_ | Hm | 22:31 |
| J1m_ | so you're getting conflicts. That's a potential downside of more clients. | 22:32 |
| goschtl | we don't use the catalog in our z3 app so i removed all from the lsm | 22:32 |
| J1m_ | Depends on your app, of course. | 22:32 |
| goschtl | now i hope all will work again | 22:32 |
| goschtl | but now i need some ideas on what i can work if the remove was not the only bottleneck | 22:33 |
| J1m_ | It's not clear if your issue is conflicts or storage throughput. I suspect the former, | 22:34 |
| goschtl | i hope it ;) | 22:35 |
| goschtl | i have all my objects organized in a balanced way | 22:35 |
| J1m_ | debugging that is harder than it should be. | 22:35 |
| goschtl | so i think it should not conflicting | 22:36 |
| goschtl | anymore | 22:36 |
| goschtl | J1m_: if i have these lots of Competing writes/reads is the a "perfomance" killer | 22:37 |
| J1m_ | It can be. | 22:59 |
| J1m_ | because you have to redo those transactions. | 23:00 |
| goschtl | thanks | 23:05 |
| *** goschtl has quit IRC | 23:06 | |
Generated by irclog2html.py 2.15.1 by Marius Gedminas - find it at mg.pov.lt!