Andrew Wong has uploaded a new change for review.

  http://gerrit.cloudera.org:8080/7654

Change subject: handle disk failures during tablet copies
......................................................................

handle disk failures during tablet copies

There are two components to tablet copies: the copy clients (that
receiving data) and the copy session sources (that sending data).

Coarse-grain handling of disk failures during tablet copies is done as
follows. For tablet copy source sessions:
- if a disk fails in the session (i.e. during a call to
  ReadFileChunkToBuf, etc.), the error should handle itself at the block
  layer and return the error to the client
- if a disk fails during the session in some other thread, the next call
  to GetBlockPiece or GetLogSegmentPiece should return the error that
  failed the replica

For tablet copy clients:
- when getting next blocks, the client repeatedly gets blocks for the
  copy. If this fails, the client will fail.
- everything will handle itself at the block layer.

Change-Id: Ic18d93c218ea13f3086f420a4847cb5e29a47bc7
---
M src/kudu/fs/data_dirs.cc
M src/kudu/tserver/tablet_copy_client.cc
M src/kudu/tserver/tablet_copy_source_session.cc
M src/kudu/tserver/ts_disk_failure-test.cc
4 files changed, 94 insertions(+), 4 deletions(-)


  git pull ssh://gerrit.cloudera.org:29418/kudu refs/changes/54/7654/1
-- 
To view, visit http://gerrit.cloudera.org:8080/7654
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: newchange
Gerrit-Change-Id: Ic18d93c218ea13f3086f420a4847cb5e29a47bc7
Gerrit-PatchSet: 1
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Andrew Wong <aw...@cloudera.com>

Reply via email to