When a tablet server (lets call it A) bulk imports a file, it makes a few bookkeeping entries in the !METADATA table. The tablet server that is serving the !METADATA table (lets call it B) checks a constraint: tablet server A must still have its zookeeper lock. This constraint is being violated because A has lost its lock.
Tablet server A should have died. The native map is used for live data ingest and exist outside of the java heap. The caches live in the heap. -Eric On Wed, Jan 15, 2014 at 8:19 AM, Anthony F <[email protected]> wrote: > Just checked on the native mem maps . . . looks like it is set to 1GB. Do > the index and data caches reside in native mem maps if available or is > native mem used for something else? > > I just repeated an ingest . . . this time I did not lose any tablet > servers but my logs are filling up with the following messages: > > 2014-01-15 08:16:41,643 [constraints.MetadataConstraints] DEBUG: violating > metadata mutation : b;74~thf > 2014-01-15 08:16:41,643 [constraints.MetadataConstraints] DEBUG: update: > file:/b-00012bq/I00012cj.rf value 20272720,0,1389757684543 > 2014-01-15 08:16:41,643 [constraints.MetadataConstraints] DEBUG: update: > loaded:/b-00012bq/I00012cj.rf value 2675766456963732003 > 2014-01-15 08:16:41,643 [constraints.MetadataConstraints] DEBUG: update: > srv:time value M1389757684543 > 2014-01-15 08:16:41,643 [constraints.MetadataConstraints] DEBUG: update: > srv:lock value tservers/ > 192.168.2.231:9997/zlock-0000000002$2438da698db13b4 > > > > On Mon, Jan 13, 2014 at 2:44 PM, Sean Busbey <[email protected]>wrote: > >> >> On Mon, Jan 13, 2014 at 12:02 PM, Anthony F <[email protected]> wrote: >> >>> Yes, system swappiness is set to 0. I'll run again and gather more logs. >>> >>> Is there a zookeeper timeout setting that I can adjust to avoid this >>> issue and is that advisable? Basically, the tservers are colocated with >>> HDFS datanodes and Hadoop nodemanagers. The machines are overallocated in >>> terms of RAM. So, I have a feeling that when a map-reduce job is kicked >>> off, it causes the tserver to page out to swap space. Once the map-reduce >>> job finishes and the bulk ingest is kicked off, the tserver is paged back >>> in and the ZK timeout causes a shutdown. >>> >>> >>> >> You should not overallocate the amount of memory on the machines. >> Generally, you should provide memory limits under teh assumption that >> everything will be on at once. >> >> Many parts of Hadoop (not just Accumulo) will degrade or malfunction in >> the presence of memory swapping. >> >> How much of hte 12GB for Accumulo is for native memmaps? >> > >
