I will change the function in my test replication. AND check the sl_event table on all servers (I never checked their size ...) Thank you for all the answers.
-----Message d'origine----- De : Steve Singer [mailto:st...@ssinger.info] Envoyé : lundi 23 juillet 2018 13:33 À : Sébastien Marchand Cc : slony1-general@lists.slony.info Objet : Re: [Slony1-general] Problème with sl_listen and too many nodes On Fri, 20 Jul 2018, Sébastien Marchand wrote: > In fact I have only 2 servers in datacenter all the others (16) are in > agency scattered in the country (low bandwidth). So no server will become > master except one. I noticed quite quickly that past a dozen nodes servers > consumes too much resource cpu, DD and network. > > So if I want the sl_listen table to no longer be filled can I modify the > function RebuildListenEntries ()? > > Thx you for your time. Yes you can modify the function. Will this break anything? I'm not sure, you would need to test. I think you might find that SYNC events generated on the agency servers with don't get deleted/removed because those events never get processed or confirmed by the other nodes. This might pose a problem with respect to sl_event growing. It would also pose a problem if slonik ever tried to perform a 'wait for event'. I don't think it would cause a problem with normal data replication. Steve > > -----Message d'origine----- > De : Steve Singer [mailto:ssin...@afilias.info] > Envoyé : jeudi 19 juillet 2018 15:33 > À : Sébastien Marchand > Cc : slony1-general@lists.slony.info > Objet : Re: [Slony1-general] Problème with sl_listen and too many nodes > > On 07/19/2018 04:05 AM, Sébastien Marchand wrote: > > RebuildListenEntries() is the function that populates sl_listen. > > With slony, any node can be the source of an event, possibly because > that event was submitted to via slonik. > > This means that node 2 must be listening for events with origin=node 2 > and origin=130. > > When you perform your 'cleanup' you actually don't have a way for node 2 > to receive events from 130. > > Node 2 could receive events directly from 130 > > origin;provider;receiver > 130,130,2 > > or via node 1 > 130,1,2 > > but here must be at least one row for each origin,receiver pair. > > If there is a sl_subscribe row connecting origin=130 to receiver=2 then > that is what the sl_listen row should also look like. > > However if there is no subscription between those nodes then the > sl_listen is built with all possibilities. I think my reasoning for > this was because if nodes fail events can still propogate(in particular > events used in the failover process). > > Is there a particular problem the larger listen network is causing? > > > > >> Sl_subscribe of my test : >> sub_set;sub_provider;sub_receiver;sub_forward;sub_active >> 1;1;2;FALSE;TRUE >> 1;1;130;FALSE;TRUE >> >> Sl_path : >> pa_server;pa_client;pa_conninfo >> 1;2;dbname=DB host=192.168.0.29 port=5432 user=slony password=123;10 >> 1;130;dbname=DB host=192.168.0.29 port=5432 user=slony password=123;10 >> 2;1;dbname=DB host=192.168.0.3 port=5432 user=slony password=123;10 >> 130;1;dbname=DB host=192.168.0.230 port=5432 user=slony password=123;10 >> >> Sub_forward true or false change nothing... >> >> Thx for your help. >> >> -----Message d'origine----- >> De : Steve Singer [mailto:st...@ssinger.info] >> Envoyé : jeudi 19 juillet 2018 05:42 >> À : Sébastien Marchand >> Cc : slony1-general@lists.slony.info >> Objet : Re: [Slony1-general] Problème with sl_listen and too many nodes >> >> On Wed, 18 Jul 2018, Sébastien Marchand wrote: >> >> What is sl_subscribe? >> >> (I assume sl_path has paths between each node) >> >> >> >> >>> >>> Hello, >>> >>> >>> >>> I have a problem with the sl_listen table. >>> >>> I have a replication that has been running for a very long time from a >> master to X slaves (1 master >>> node for 18 nodes) >>> >>> My concern comes from the sl_listen table which instead of just containing >> what it needs it creates >>> all the possible combinations. >>> >>> For example I have a test for 1 master and 2 slaves and I have 4 too many >> lines that are useless >>> for me: >>> >>> >>> >>> SL_LISTEN table >>> >>> origin; provider; receiver >>> >>> 1, 1, 2 >>> >>> 1; 1; 130 >>> >>> 2; 1; 130 >>> >>> 2; 2; 1 >>> >>> 2; 130; 1 >>> >>> 130, 1, 2 >>> >>> 130, 2, 1 >>> >>> 130; 130; 1 >>> >>> >>> >>> Here after cleaning what should be: >>> >>> >>> >>> origin; provider; receiver >>> >>> 1, 1, 2 >>> >>> 1; 1; 130 >>> >>> 2; 2; 1 >>> >>> 130; 130; 1 >>> >>> >>> >>> The problem is that with each add / delete of tables / nodes the table is >> re-filled again and I >>> have to redo the cleaning. >>> >>> >>> >>> My final question is: Is it normal for all nodes to talk to each other? >>> >>> >>> >>> Thank you. >>> >>> >>> >> >> _______________________________________________ >> Slony1-general mailing list >> Slony1-general@lists.slony.info >> http://lists.slony.info/mailman/listinfo/slony1-general >> > > > _______________________________________________ > Slony1-general mailing list > Slony1-general@lists.slony.info > http://lists.slony.info/mailman/listinfo/slony1-general _______________________________________________ Slony1-general mailing list Slony1-general@lists.slony.info http://lists.slony.info/mailman/listinfo/slony1-general
