|
 |
21ab4e |
From c58fcc9cfff6d7858febcc53847cda7624c2d982 Mon Sep 17 00:00:00 2001
|
|
|
21ab4e |
From: Ashish Pandey <aspandey@redhat.com>
|
|
|
21ab4e |
Date: Thu, 22 Jun 2017 17:06:40 +0530
|
|
|
21ab4e |
Subject: [PATCH 532/539] ec: Increase notification in all the cases
|
|
|
21ab4e |
|
|
|
21ab4e |
Problem:
|
|
|
21ab4e |
"gluster v heal <volname> info" is taking
|
|
|
21ab4e |
long time to respond when a brick is down.
|
|
|
21ab4e |
|
|
|
21ab4e |
RCA:
|
|
|
21ab4e |
Heal info command does virtual mount.
|
|
|
21ab4e |
EC wait for 10 seconds, before sending UP call to upper xlator,
|
|
|
21ab4e |
to get notification (DOWN or UP) from all the bricks.
|
|
|
21ab4e |
|
|
|
21ab4e |
Currently, we are increasing ec->xl_notify_count based on
|
|
|
21ab4e |
the current status of the brick. So, if a DOWN event notification
|
|
|
21ab4e |
has come and brick is already down, we are not increasing
|
|
|
21ab4e |
ec->xl_notify_count in ec_handle_down.
|
|
|
21ab4e |
|
|
|
21ab4e |
Solution:
|
|
|
21ab4e |
Handle DOWN even as notification irrespective of what
|
|
|
21ab4e |
is the current status of brick.
|
|
|
21ab4e |
|
|
|
21ab4e |
>Change-Id: I0acac0db7ec7622d4c0584692e88ad52f45a910f
|
|
|
21ab4e |
>BUG: 1464091
|
|
|
21ab4e |
>Signed-off-by: Ashish Pandey <aspandey@redhat.com>
|
|
|
21ab4e |
>Reviewed-on: https://review.gluster.org/17606
|
|
|
21ab4e |
>Tested-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
21ab4e |
>Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
21ab4e |
>Smoke: Gluster Build System <jenkins@build.gluster.org>
|
|
|
21ab4e |
>CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
|
|
|
21ab4e |
>Reviewed-by: Xavier Hernandez <xhernandez@datalab.es>
|
|
|
21ab4e |
>NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
|
|
|
21ab4e |
>Signed-off-by: Ashish Pandey <aspandey@redhat.com>
|
|
|
21ab4e |
|
|
|
21ab4e |
Change-Id: I0acac0db7ec7622d4c0584692e88ad52f45a910f
|
|
|
21ab4e |
BUG: 1463108
|
|
|
21ab4e |
Signed-off-by: Ashish Pandey <aspandey@redhat.com>
|
|
|
21ab4e |
Reviewed-on: https://code.engineering.redhat.com/gerrit/110286
|
|
|
21ab4e |
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
21ab4e |
---
|
|
|
21ab4e |
xlators/cluster/ec/src/ec.c | 51 ++++++++++++++++++---------------------------
|
|
|
21ab4e |
1 file changed, 20 insertions(+), 31 deletions(-)
|
|
|
21ab4e |
|
|
|
21ab4e |
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
|
|
|
21ab4e |
index bad5578..f1aeea7 100644
|
|
|
21ab4e |
--- a/xlators/cluster/ec/src/ec.c
|
|
|
21ab4e |
+++ b/xlators/cluster/ec/src/ec.c
|
|
|
21ab4e |
@@ -401,35 +401,6 @@ ec_launch_notify_timer (xlator_t *this, ec_t *ec)
|
|
|
21ab4e |
}
|
|
|
21ab4e |
}
|
|
|
21ab4e |
|
|
|
21ab4e |
-void
|
|
|
21ab4e |
-ec_handle_up (xlator_t *this, ec_t *ec, int32_t idx)
|
|
|
21ab4e |
-{
|
|
|
21ab4e |
- if (((ec->xl_up >> idx) & 1) == 0) { /* Duplicate event */
|
|
|
21ab4e |
- if (((ec->xl_notify >> idx) & 1) == 0) {
|
|
|
21ab4e |
- ec->xl_notify |= 1ULL << idx;
|
|
|
21ab4e |
- ec->xl_notify_count++;
|
|
|
21ab4e |
- }
|
|
|
21ab4e |
- ec->xl_up |= 1ULL << idx;
|
|
|
21ab4e |
- ec->xl_up_count++;
|
|
|
21ab4e |
- }
|
|
|
21ab4e |
-}
|
|
|
21ab4e |
-
|
|
|
21ab4e |
-void
|
|
|
21ab4e |
-ec_handle_down (xlator_t *this, ec_t *ec, int32_t idx)
|
|
|
21ab4e |
-{
|
|
|
21ab4e |
- if (((ec->xl_up >> idx) & 1) != 0) { /* Duplicate event */
|
|
|
21ab4e |
- gf_msg_debug (this->name, 0, "Child %d is DOWN", idx);
|
|
|
21ab4e |
-
|
|
|
21ab4e |
- if (((ec->xl_notify >> idx) & 1) == 0) {
|
|
|
21ab4e |
- ec->xl_notify |= 1ULL << idx;
|
|
|
21ab4e |
- ec->xl_notify_count++;
|
|
|
21ab4e |
- }
|
|
|
21ab4e |
-
|
|
|
21ab4e |
- ec->xl_up ^= 1ULL << idx;
|
|
|
21ab4e |
- ec->xl_up_count--;
|
|
|
21ab4e |
- }
|
|
|
21ab4e |
-}
|
|
|
21ab4e |
-
|
|
|
21ab4e |
gf_boolean_t
|
|
|
21ab4e |
ec_disable_delays(ec_t *ec)
|
|
|
21ab4e |
{
|
|
|
21ab4e |
@@ -446,6 +417,22 @@ ec_pending_fops_completed(ec_t *ec)
|
|
|
21ab4e |
}
|
|
|
21ab4e |
}
|
|
|
21ab4e |
|
|
|
21ab4e |
+static void
|
|
|
21ab4e |
+ec_set_up_state(ec_t *ec, uintptr_t index_mask, uintptr_t new_state)
|
|
|
21ab4e |
+{
|
|
|
21ab4e |
+ uintptr_t current_state = 0;
|
|
|
21ab4e |
+
|
|
|
21ab4e |
+ if ((ec->xl_notify & index_mask) == 0) {
|
|
|
21ab4e |
+ ec->xl_notify |= index_mask;
|
|
|
21ab4e |
+ ec->xl_notify_count++;
|
|
|
21ab4e |
+ }
|
|
|
21ab4e |
+ current_state = ec->xl_up & index_mask;
|
|
|
21ab4e |
+ if (current_state != new_state) {
|
|
|
21ab4e |
+ ec->xl_up ^= index_mask;
|
|
|
21ab4e |
+ ec->xl_up_count += (current_state ? -1 : 1);
|
|
|
21ab4e |
+ }
|
|
|
21ab4e |
+}
|
|
|
21ab4e |
+
|
|
|
21ab4e |
int32_t
|
|
|
21ab4e |
ec_notify (xlator_t *this, int32_t event, void *data, void *data2)
|
|
|
21ab4e |
{
|
|
|
21ab4e |
@@ -459,6 +446,7 @@ ec_notify (xlator_t *this, int32_t event, void *data, void *data2)
|
|
|
21ab4e |
int32_t orig_event = event;
|
|
|
21ab4e |
struct gf_upcall *up_data = NULL;
|
|
|
21ab4e |
struct gf_upcall_cache_invalidation *up_ci = NULL;
|
|
|
21ab4e |
+ uintptr_t mask = 0;
|
|
|
21ab4e |
|
|
|
21ab4e |
gf_msg_trace (this->name, 0, "NOTIFY(%d): %p, %p",
|
|
|
21ab4e |
event, data, data2);
|
|
|
21ab4e |
@@ -510,10 +498,11 @@ ec_notify (xlator_t *this, int32_t event, void *data, void *data2)
|
|
|
21ab4e |
if (idx < ec->nodes) { /* CHILD_* events */
|
|
|
21ab4e |
old_event = ec_get_event_from_state (ec);
|
|
|
21ab4e |
|
|
|
21ab4e |
+ mask = 1ULL << idx;
|
|
|
21ab4e |
if (event == GF_EVENT_CHILD_UP) {
|
|
|
21ab4e |
- ec_handle_up (this, ec, idx);
|
|
|
21ab4e |
+ ec_set_up_state(ec, mask, mask);
|
|
|
21ab4e |
} else if (event == GF_EVENT_CHILD_DOWN) {
|
|
|
21ab4e |
- ec_handle_down (this, ec, idx);
|
|
|
21ab4e |
+ ec_set_up_state(ec, mask, 0);
|
|
|
21ab4e |
}
|
|
|
21ab4e |
|
|
|
21ab4e |
event = ec_get_event_from_state (ec);
|
|
|
21ab4e |
--
|
|
|
21ab4e |
1.8.3.1
|
|
|
21ab4e |
|