diff options
| author | Richard Wareing <rwareing@fb.com> | 2015-09-25 13:42:40 -0700 |
|---|---|---|
| committer | Jeff Darcy <jdarcy@fb.com> | 2017-07-03 07:48:08 -0700 |
| commit | 992a9f8494a358f828eeef34b46e9f5ccfca1d3b (patch) | |
| tree | 353da3de0f4a72f5ddfaaa07778e85cb4a232ef0 /xlators/cluster/afr/src | |
| parent | 32fc893c176965109d58a3398ad042683d159d77 (diff) | |
cluster/afr: Fix metadata split-brain flow (HOTFIX)
Summary:
- The metadata heal flow for some reason likes to tinker with the sink
states prior to having the source finalized, this broke the
policy based unsplit flow. This patch fixes it by simply setting
those chilren who aren't the favorite as sinks.
Test Plan: - Tested against some reported instances
Reviewers: moox, sshreyas, dph
Reviewed By: sshreyas
Differential Revision: https://phabricator.fb.com/D2481527
Signature: t1:2481527:1443215555:1165d8eb5f3dec216ec3ff0795d9837712906b1d
Blame Revision: Change-Id: I56f96fdcef32dd4fc5d35958148d0e56d142d5e4
Change-Id: I16aa445a22c3bcd7b589954e2da513ed53822d5b
Diffstat (limited to 'xlators/cluster/afr/src')
| -rw-r--r-- | xlators/cluster/afr/src/afr-self-heal-metadata.c | 17 |
1 files changed, 17 insertions, 0 deletions
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c index d5d95ef8852..44374d5c0ad 100644 --- a/xlators/cluster/afr/src/afr-self-heal-metadata.c +++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c @@ -216,6 +216,23 @@ __afr_selfheal_metadata_finalize_source (call_frame_t *frame, xlator_t *this, sources_count = AFR_COUNT (sources, priv->child_count); + if (sources_count == 0) { + afr_sh_get_source_by_policy (this, sources, healed_sinks, + locked_on, replies, inode); + sources_count = AFR_COUNT (sources, priv->child_count); + /* __afr_selfheal_metadata_prepare tinkers with the state + * of healed_sinks pre-maturely (the source hasn't + * actually been finalized yet!), so reset the children + * which aren't our source to sinks so we can heal. + * I'll leave it to the AFR2 maintainer to fix that code + * in the future as they may have had a good reason. + */ + for (i = 0; i < priv->child_count; i++) { + if (!sources[i] && locked_on[i]) + healed_sinks[i] = 1; + } + } + if ((AFR_CMP (locked_on, healed_sinks, priv->child_count) == 0) || !sources_count) { |