I'm having a simmular problem, but i get verry random read problems 4 bytes timouts. (mailed before about this)
The only fix i got/found was, switch to framed transport instead of buffered transport, but i still have the same problem (less frequent) running debian packaged cassandra (0.6.1) with phpcassa lib (using thrift) I have it localhost and external host, so network/switching is not the problem.. I'm afraid there is still something wrong with php thrift.. (even after patch https://issues.apache.org/jira/browse/THRIFT-347) PS. i see there is 0.6.3 now available for debian, I'll test that version.. like you said you don't have the issue on 0.6.3 it might be fixed there.. Pieter Maes Op 6/07/10 10:14, Juho Mäkinen schreef: > I'm having a strange problem when some of get_column and get_slice calls > fails to "TSocket: Could not write 85 bytes paja.dev.pri:9160" exception. > The problem appears on some but not all calls to get_column and also > some times to other calls. > > I'm using 0.6.3 on a single host (it's a development installation) but > the problem appeared on 0.6.1. The client side is done with php using > the thrift api. > > I've turned on TRACE level debugging on the cassandra server and the requests > which doesn't fail produce this kind of debug to the logs: > > DEBUG [pool-1-thread-4] 2010-07-06 11:12:23,438 CassandraServer.java > (line 266) get > DEBUG [ROW-READ-STAGE:2] 2010-07-06 11:12:23,439 StorageProxy.java > (line 755) weakreadlocal reading SliceByNamesReadCommand(table='ig', > key='2249', columnParent='QueryPath(columnFamilyName='TalkLastMessages', > superColumnName='null', columnName='null')', columns=[2582837,]) > > But the requests which fails to the TSocket: Could not write 85 bytes > -exception > doesn't print anything to the logs. I'm doing the same request all the > time with same > parameters. > > How I should debug this issue? All ideas and tips are greatly appreciated. > > - Juho Mäkinen
