diff options
| author | Brian Behlendorf <behlendorf1@llnl.gov> | 2022-05-24 16:36:07 +0000 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2022-05-24 16:36:07 +0000 |
| commit | cf70c0f8ae01c24699767b5ecfbe1882c009d53a (patch) | |
| tree | 9790515a7a2701c7aeadf3c4a5e89e8c81ae9aa8 | |
| parent | 2cd0f98f4aae7110a48cb0623e1e3d66b9f58785 (diff) | |
| download | src-cf70c0f8ae01c24699767b5ecfbe1882c009d53a.tar.gz src-cf70c0f8ae01c24699767b5ecfbe1882c009d53a.zip | |
zed: Take no action on scrub/resilver checksum errors
When scrubbing/resilvering a pool it can be counter productive to
cancel the scan and kick of a replace operation to a hot spare
when encountering checksum errors. In this case, the best course
of action is to allow the scrub/resilver to complete as quickly
as possible and to keep the vdevs fully online if possible.
Realistically, this is less of an issue for a RAIDZ since a
traditional resilver must be used and checksums will be verified.
However, this is not the case for a mirror or dRAID pool which is
sequentially resilvered and checksum verification is deferred
until after the replace operation completes.
Regardless, we apply this policy to all pool types since it's
a good idea for all vdevs. Degrading additional vdevs has the
potential to make a bad situation worse. Note the checksum
errors will still be reported as both an event and by
`zpool status`. This change only prevents the ZED from
proactively taking any action.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #13499
| -rw-r--r-- | cmd/zed/agents/zfs_diagnosis.c | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/cmd/zed/agents/zfs_diagnosis.c b/cmd/zed/agents/zfs_diagnosis.c index 4fe9ee2bc5b2..813916d2e82c 100644 --- a/cmd/zed/agents/zfs_diagnosis.c +++ b/cmd/zed/agents/zfs_diagnosis.c @@ -34,6 +34,7 @@ #include <sys/fs/zfs.h> #include <sys/fm/protocol.h> #include <sys/fm/fs/zfs.h> +#include <sys/zio.h> #include "zfs_agents.h" #include "fmd_api.h" @@ -770,6 +771,8 @@ zfs_fm_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl, const char *class) ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_PROBE_FAILURE))) { char *failmode = NULL; boolean_t checkremove = B_FALSE; + uint32_t pri = 0; + int32_t flags = 0; /* * If this is a checksum or I/O error, then toss it into the @@ -792,6 +795,23 @@ zfs_fm_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl, const char *class) checkremove = B_TRUE; } else if (fmd_nvl_class_match(hdl, nvl, ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_CHECKSUM))) { + /* + * We ignore ereports for checksum errors generated by + * scrub/resilver I/O to avoid potentially further + * degrading the pool while it's being repaired. + */ + if (((nvlist_lookup_uint32(nvl, + FM_EREPORT_PAYLOAD_ZFS_ZIO_PRIORITY, &pri) == 0) && + (pri == ZIO_PRIORITY_SCRUB || + pri == ZIO_PRIORITY_REBUILD)) || + ((nvlist_lookup_int32(nvl, + FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS, &flags) == 0) && + (flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER)))) { + fmd_hdl_debug(hdl, "ignoring '%s' for " + "scrub/resilver I/O", class); + return; + } + if (zcp->zc_data.zc_serd_checksum[0] == '\0') { zfs_serd_name(zcp->zc_data.zc_serd_checksum, pool_guid, vdev_guid, "checksum"); |