On Tue, May 24, 2011 at 12:40 AM, Daniel Doubleday <daniel.double...@gmx.net> wrote: > We are performing the repair on one node only. Other nodes receive reasonable > amounts of data (~500MB). It's only the repairing node itself which > 'explodes'.
That, for instance, is a bit weird. That the node on which the repair is performed get more data is expected, since it is repair with all it's "neighbor" while the neighbors themselves get repaired only against that given node. But when differences between two A and B are computed, the ranges to repair are streaming both from A to B and for B to A. Unless A and B are widely out of sync (like A has no data and B has tons of it), around the same amount of data should transit in both way. So with RF=3, the node on with repair was started should get around 4 times (up to 6 times if you have weird topology) as much data than any neighboring node, but that's is. While if I'm correct, you are reporting that the neighboring node gets ~500MB and the "coordinator" gets > 700GB ?! Honestly I'm not sure an imprecision of the merkle tree could account for that behavior. Anyway, Daniel, would you be able to share the logs of the nodes (at least the node on which repair is started) ? I'm not sure how much that could help but that cannot hurt. -- Sylvain > > I must admit that I'm a noob when it comes to aes/repair. Its just strange > that a cluster that is up and running with no probs is doing that. But I > understand that its not supposed to do what its doing. I just hope that I > find out why soon enough. > > > > On 23.05.2011, at 21:21, Peter Schuller <peter.schul...@infidyne.com> wrote: > >>> I'm a bit lost: I tried a repair yesterday with only one CF and that didn't >>> really work the way I expected but I thought that would be a bug which only >>> affects that special case. >>> >>> So I tried again for all CFs. >>> >>> I started with a nicely compacted machine with around 320GB of load. Total >>> disc space on this node was 1.1TB. >> >> Did you do repairs simultaneously on all nodes? >> >> I have seen very significant disk space increases under some >> circumstances. While I haven't filed a ticket about it because there >> was never time to confirm, I believe two things were at play: >> >> (1) nodes were sufficiently out a sync in a sufficiently spread out >> fashion that the granularity of the merkle tree (IIRC, and if I read >> correctly, it divides the ring into up to 2^15 segments but no more) >> became ineffective so that repair effectively had to transfer all the >> data. at first I thought there was an outright bug, but after looking >> at the code I suspected it was just the merkle tree granularity. >> >> (2) I suspected at the time that a contributing factor was also that >> as one repair might cause a node to significantly increase it's live >> sstables temporarily until they are compacted, another repair on >> another node may start and start validating compaction and streaming >> of that data - leading to disk space bload essentially being >> "contagious"; the third node streaming from the node that was >> temporarily bloated, will receive even more data from that node than >> it normally would. >> >> We're making sure to only run one repair at a time between any hosts >> that are neighbors of each other (meaning that at RF=3, that's 1 >> concurrent repair per 6 nodes in the cluster). >> >> I'd be interested in hearing anyone confirm or deny whether my >> understanding of (1) in particular is correct. To connect it to >> reality: a 20 GB CF divided into 2^15 segments implies each segment is >>> 600 kbyte in size. For CF:s with tens or hundreds of millions of >> small rows and a fairly random (with respect to partitioner) update >> pattern, it's not very difficult to end up in a situation where most >> 600 kbyte chunks contain out-of-synch data. Particularly in a >> situation with lots of dropped messages. >> >> I'm getting the 2^15 from AntiEntropyService.Validator.Validator() >> which passes a maxsize of 2^15 to the MerkelTree constructor. >> >> -- >> / Peter Schuller >