[1/2] kudu git commit: KUDU-2052: use XFS_IOC_UNRESVSP64 ioctl to punch holes on xfs

2017-06-23 Thread adar
Repository: kudu Updated Branches: refs/heads/master 410763afa -> c96d6680b KUDU-2052: use XFS_IOC_UNRESVSP64 ioctl to punch holes on xfs There's not much to say beyond what's already in the bug report. To test, I created a 128 GB loopback-mounted xfs filesystem and ran the new benchmark on

[2/2] kudu git commit: Revert "KUDU-2052: disable hole repunching at startup on xfs filesystems"

2017-06-23 Thread adar
Revert "KUDU-2052: disable hole repunching at startup on xfs filesystems" With the proper fix for KUDU-2052 ready, we no longer need this workaround. This reverts commit 410763afa61252fa8c7125f66429bad602db795a. Change-Id: Ie36750f419574441c6e5d9dd10a0558eef8cdbe9 Reviewed-on:

[1/2] kudu git commit: fs: add filesystem type detection to data dirs

2017-06-23 Thread adar
Repository: kudu Updated Branches: refs/heads/master 44bd94268 -> 410763afa fs: add filesystem type detection to data dirs This patch moves some detection code out of LogBlockManager and into DataDirManager. It also adds detection of xfs filesystems. To verify that I didn't regress the

[2/2] kudu git commit: KUDU-2052: disable hole repunching at startup on xfs filesystems

2017-06-23 Thread adar
KUDU-2052: disable hole repunching at startup on xfs filesystems As a quick and dirty workaround for (what we believe to be) the root cause of KUDU-2052, let's disable hole repunching at startup for containers that reside on xfs filesystems. This toggle is gated behind a gflag which is disabled

kudu git commit: [raft_consensus-itest] disable TestHammerOneRow for a while

2017-06-23 Thread alexey
Repository: kudu Updated Branches: refs/heads/master 5121df65e -> 44bd94268 [raft_consensus-itest] disable TestHammerOneRow for a while Changelist c25d127ad3512ed9c9a2565681ffce9de639639f exposed inconsistent verification of SetFlag() results in the EnableLogLatency() function. This patch

kudu git commit: KUDU-1955 refuse to use world-readable keytabs

2017-06-23 Thread danburkert
Repository: kudu Updated Branches: refs/heads/master f9e51ca63 -> 5121df65e KUDU-1955 refuse to use world-readable keytabs Allowing users to supply keytab files and TLS private keys with world-readable permissions lessens a cluster's security. During Kerberos/TLS initialization, servers now

kudu git commit: KUDU-2049 Fix too-strict CHECK in RleIntBlockDecoder::SeekToPositionInBlock

2017-06-23 Thread wdberkeley
Repository: kudu Updated Branches: refs/heads/master c25d127ad -> f9e51ca63 KUDU-2049 Fix too-strict CHECK in RleIntBlockDecoder::SeekToPositionInBlock In a CFile block with n non-null values, it's permissible to seek to the spot after all rows. However, a CHECK_LT in

kudu git commit: Fix SIGSEGV in ksck

2017-06-23 Thread dralves
Repository: kudu Updated Branches: refs/heads/master d62394293 -> def06d8b8 Fix SIGSEGV in ksck ksck will segfault when some tablet servers that host tablet replicas are missing. This happens, for example, if the master is still restarting and has not yet fully populated its list of live