|
 |
d1681e |
From 87a24e342c422ba6b04563d63d431430c0156b52 Mon Sep 17 00:00:00 2001
|
|
|
d1681e |
From: N Balachandran <nbalacha@redhat.com>
|
|
|
d1681e |
Date: Fri, 6 Apr 2018 16:06:51 +0530
|
|
|
d1681e |
Subject: [PATCH 217/236] cluster/dht: Handle file migrations when brick down
|
|
|
d1681e |
|
|
|
d1681e |
The decision as to which node would migrate a file
|
|
|
d1681e |
was based on the gfid of the file. Files were divided
|
|
|
d1681e |
among the nodes for the replica/disperse set. However,
|
|
|
d1681e |
if a brick was down when rebalance started, the nodeuuids
|
|
|
d1681e |
would be saved as NULL and a set of files would not be migrated.
|
|
|
d1681e |
|
|
|
d1681e |
Now, if the nodeuuid is NULL, the first non-null entry in
|
|
|
d1681e |
the set is the node responsible for migrating the file.
|
|
|
d1681e |
|
|
|
d1681e |
upstream master: https://review.gluster.org/#/c/19831/
|
|
|
d1681e |
|
|
|
d1681e |
> Change-Id: I72554c107792c7d534e0f25640654b6f8417d373
|
|
|
d1681e |
> fixes: bz#1564198
|
|
|
d1681e |
> Signed-off-by: N Balachandran <nbalacha@redhat.com>
|
|
|
d1681e |
|
|
|
d1681e |
Change-Id: Ia0e15339aefee2712e85d7e282c9b7934665376b
|
|
|
d1681e |
BUG: 1553677
|
|
|
d1681e |
Signed-off-by: N Balachandran <nbalacha@redhat.com>
|
|
|
d1681e |
Reviewed-on: https://code.engineering.redhat.com/gerrit/135515
|
|
|
d1681e |
Tested-by: RHGS Build Bot <nigelb@redhat.com>
|
|
|
d1681e |
Reviewed-by: Atin Mukherjee <amukherj@redhat.com>
|
|
|
d1681e |
---
|
|
|
d1681e |
xlators/cluster/dht/src/dht-rebalance.c | 56 ++++++++++++++++++++++++++++++---
|
|
|
d1681e |
1 file changed, 51 insertions(+), 5 deletions(-)
|
|
|
d1681e |
|
|
|
d1681e |
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
|
|
|
d1681e |
index bba44b9..a4be348 100644
|
|
|
d1681e |
--- a/xlators/cluster/dht/src/dht-rebalance.c
|
|
|
d1681e |
+++ b/xlators/cluster/dht/src/dht-rebalance.c
|
|
|
d1681e |
@@ -2469,6 +2469,27 @@ gf_defrag_ctx_subvols_init (dht_dfoffset_ctx_t *offset_var, xlator_t *this) {
|
|
|
d1681e |
}
|
|
|
d1681e |
|
|
|
d1681e |
|
|
|
d1681e |
+static int
|
|
|
d1681e |
+dht_get_first_non_null_index (subvol_nodeuuids_info_t *entry)
|
|
|
d1681e |
+{
|
|
|
d1681e |
+ int i = 0;
|
|
|
d1681e |
+ int index = 0;
|
|
|
d1681e |
+
|
|
|
d1681e |
+ for (i = 0; i < entry->count; i++) {
|
|
|
d1681e |
+ if (!gf_uuid_is_null (entry->elements[i].uuid)) {
|
|
|
d1681e |
+ index = i;
|
|
|
d1681e |
+ goto out;
|
|
|
d1681e |
+ }
|
|
|
d1681e |
+ }
|
|
|
d1681e |
+
|
|
|
d1681e |
+ if (i == entry->count) {
|
|
|
d1681e |
+ index = -1;
|
|
|
d1681e |
+ }
|
|
|
d1681e |
+out:
|
|
|
d1681e |
+ return index;
|
|
|
d1681e |
+}
|
|
|
d1681e |
+
|
|
|
d1681e |
+
|
|
|
d1681e |
/* Return value
|
|
|
d1681e |
* 0 : this node does not migrate the file
|
|
|
d1681e |
* 1 : this node migrates the file
|
|
|
d1681e |
@@ -2485,28 +2506,53 @@ gf_defrag_should_i_migrate (xlator_t *this, int local_subvol_index, uuid_t gfid)
|
|
|
d1681e |
int i = local_subvol_index;
|
|
|
d1681e |
char *str = NULL;
|
|
|
d1681e |
uint32_t hashval = 0;
|
|
|
d1681e |
- int32_t index = 0;
|
|
|
d1681e |
+ int32_t index = 0;
|
|
|
d1681e |
dht_conf_t *conf = NULL;
|
|
|
d1681e |
char buf[UUID_CANONICAL_FORM_LEN + 1] = {0, };
|
|
|
d1681e |
+ subvol_nodeuuids_info_t *entry = NULL;
|
|
|
d1681e |
+
|
|
|
d1681e |
|
|
|
d1681e |
conf = this->private;
|
|
|
d1681e |
|
|
|
d1681e |
- /* Pure distribute */
|
|
|
d1681e |
+ /* Pure distribute. A subvol in this case
|
|
|
d1681e |
+ will be handled by only one node */
|
|
|
d1681e |
|
|
|
d1681e |
- if (conf->local_nodeuuids[i].count == 1) {
|
|
|
d1681e |
+ entry = &(conf->local_nodeuuids[i]);
|
|
|
d1681e |
+ if (entry->count == 1) {
|
|
|
d1681e |
return 1;
|
|
|
d1681e |
}
|
|
|
d1681e |
|
|
|
d1681e |
str = uuid_utoa_r (gfid, buf);
|
|
|
d1681e |
ret = dht_hash_compute (this, 0, str, &hashval);
|
|
|
d1681e |
if (ret == 0) {
|
|
|
d1681e |
- index = (hashval % conf->local_nodeuuids[i].count);
|
|
|
d1681e |
- if (conf->local_nodeuuids[i].elements[index].info
|
|
|
d1681e |
+ index = (hashval % entry->count);
|
|
|
d1681e |
+ if (entry->elements[index].info
|
|
|
d1681e |
== REBAL_NODEUUID_MINE) {
|
|
|
d1681e |
/* Index matches this node's nodeuuid.*/
|
|
|
d1681e |
ret = 1;
|
|
|
d1681e |
+ goto out;
|
|
|
d1681e |
+ }
|
|
|
d1681e |
+
|
|
|
d1681e |
+ /* Brick down - some other node has to migrate these files*/
|
|
|
d1681e |
+ if (gf_uuid_is_null (entry->elements[index].uuid)) {
|
|
|
d1681e |
+ /* Fall back to the first non-null index */
|
|
|
d1681e |
+ index = dht_get_first_non_null_index (entry);
|
|
|
d1681e |
+
|
|
|
d1681e |
+ if (index == -1) {
|
|
|
d1681e |
+ /* None of the bricks in the subvol are up.
|
|
|
d1681e |
+ * CHILD_DOWN will kill the process soon */
|
|
|
d1681e |
+
|
|
|
d1681e |
+ return 0;
|
|
|
d1681e |
+ }
|
|
|
d1681e |
+
|
|
|
d1681e |
+ if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
|
|
|
d1681e |
+ /* Index matches this node's nodeuuid.*/
|
|
|
d1681e |
+ ret = 1;
|
|
|
d1681e |
+ goto out;
|
|
|
d1681e |
+ }
|
|
|
d1681e |
}
|
|
|
d1681e |
}
|
|
|
d1681e |
+out:
|
|
|
d1681e |
return ret;
|
|
|
d1681e |
}
|
|
|
d1681e |
|
|
|
d1681e |
--
|
|
|
d1681e |
1.8.3.1
|
|
|
d1681e |
|