To be a bit more precise, we have 375GB of data. Still, I am under the impression that 1.4.2 uses at least 3 times more RAM than 1.3.4. Everything was fine under 1.3.4.
On Friday, December 26, 2014 11:52:48 PM UTC+3, Jean-Noël Rivasseau wrote: > > Hello, > > We had a 2 nodes, 10 shards cluster with about 150GB of data (not sure how > many documents, I would say about 150 millions since 1 doc is about 1K for > us). It worked fine under ES 1.3.4. > > We tried to update it today to 1.4.2. Now the nodes won't start. They eat > up all RAM during the recovery phase (when booting, and starting engines on > all indices), then GC occurs when no more RAM is available for ES. ES then > eventually gets stuck with OOM errors. > > I tried to increase the RAM given to ES to 30GB and it still eats all that > up and fails. Before with 1.3.4 we had 16GB allocated and no problems. > > Why ES needs all that RAM? It's doing nothing, not servicing a single > request... it's only starting the recovery process... > > We now have our production cluster down. It's really a huge problem for > us, please advise on any solutions or things we could try out. > > Java is IcedTea 7.2.5.3 > OS is Gentoo Linux > -- You received this message because you are subscribed to the Google Groups "elasticsearch" group. To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/elasticsearch/8a769d4b-65b7-4809-93a9-86b36e64e459%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.