cluster/ec: avoid delays in self-heal

Self-heal creates a thread per brick to sweep the index looking for
files that need to be healed. These threads are started before the
volume comes online, so nothing is done but waiting for the next
sweep. This happens once per minute.

When a replace brick command is executed, the new graph is loaded and
all index sweeper threads started. When all bricks have reported, a
getxattr request is sent to the root directory of the volume. This
causes a heal on it (because the new brick doesn't have good data),
and marks its contents as pending to be healed. This is done by the
index sweeper thread on the next round, one minute later.

This patch solves this problem by waking all index sweeper threads
after a successful check on the root directory.

Additionally, the index sweep thread scans the index directory
sequentially, but it might happen that after healing a directory entry
more index entries are created but skipped by the current directory
scan. This causes the remaining entries to be processed on the next
round, one minute later. The same can happen in the next round, so
the heal is running in bursts and taking a lot to finish, specially
on volumes with many directory levels.

This patch solves this problem by immediately restarting the index
sweep if a directory has been healed.

Change-Id: I58d9ab6ef17b30f704dc322e1d3d53b904e5f30e
BUG: 1547662
Signed-off-by: Xavi Hernandez <jahernan@redhat.com>

4 files changed, 93 insertions, 48 deletions

diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c
index 6562adf9e24..d1e40607e33 100644
--- a/xlators/cluster/ec/src/ec-heal.c
+++ b/xlators/cluster/ec/src/ec-heal.c
@@ -25,6 +25,7 @@
 #include "ec-combine.h"
 #include "ec-method.h"
 #include "ec-fops.h"
+#include "ec-heald.h"
 
 #define alloca0(size) ({void *__ptr; __ptr = alloca(size); memset(__ptr, 0, size); __ptr; })
 #define EC_COUNT(array, max) ({int __i; int __res = 0; for (__i = 0; __i < max; __i++) if (array[__i]) __res++; __res; })
@@ -2769,6 +2770,14 @@ ec_replace_heal (ec_t *ec, inode_t *inode)
                 gf_msg_debug (ec->xl->name, 0,
                         "Heal failed for replace brick ret = %d", ret);
 
+        /* Once the root inode has been checked, it might have triggered a
+         * self-heal on it after a replace brick command or for some other
+         * reason. It can also happen that the volume already had damaged
+         * files in the index, even if the heal on the root directory failed.
+         * In both cases we need to wake all index healers to continue
+         * healing remaining entries that are marked as dirty. */
+        ec_shd_index_healer_wake(ec);
+
         loc_wipe (&loc);
         return ret;
 }
diff --git a/xlators/cluster/ec/src/ec-heald.c b/xlators/cluster/ec/src/ec-heald.c
index b4fa6f87189..a703379a59b 100644
--- a/xlators/cluster/ec/src/ec-heald.c
+++ b/xlators/cluster/ec/src/ec-heald.c
@@ -184,8 +184,19 @@ ec_shd_index_purge (xlator_t *subvol, inode_t *inode, char *name)
 int
 ec_shd_selfheal (struct subvol_healer *healer, int child, loc_t *loc)
 {
-        return syncop_getxattr (healer->this, loc, NULL, EC_XATTR_HEAL, NULL,
-                                NULL);
+        int32_t ret;
+
+        ret = syncop_getxattr (healer->this, loc, NULL, EC_XATTR_HEAL, NULL,
+                               NULL);
+        if ((ret >= 0) && (loc->inode->ia_type == IA_IFDIR)) {
+                /* If we have just healed a directory, it's possible that
+                 * other index entries have appeared to be healed. We put a
+                 * mark so that we can check it later and restart a scan
+                 * without delay. */
+                healer->rerun = _gf_true;
+        }
+
+        return ret;
 }
 
 
@@ -472,11 +483,15 @@ ec_shd_index_healer_spawn (xlator_t *this, int subvol)
 }
 
 void
-ec_selfheal_childup (ec_t *ec, int child)
+ec_shd_index_healer_wake(ec_t *ec)
 {
-        if (!ec->shd.iamshd)
-                return;
-        ec_shd_index_healer_spawn (ec->xl, child);
+        int32_t i;
+
+        for (i = 0; i < ec->nodes; i++) {
+                if (((ec->xl_up >> i) & 1) != 0) {
+                        ec_shd_index_healer_spawn(ec->xl, i);
+                }
+        }
 }
 
 int
diff --git a/xlators/cluster/ec/src/ec-heald.h b/xlators/cluster/ec/src/ec-heald.h
index 4ae02e2df3c..2a8488124c4 100644
--- a/xlators/cluster/ec/src/ec-heald.h
+++ b/xlators/cluster/ec/src/ec-heald.h
@@ -20,6 +20,8 @@ ec_xl_op (xlator_t *this, dict_t *input, dict_t *output);
 
 int
 ec_selfheal_daemon_init (xlator_t *this);
-void ec_selfheal_childup (ec_t *ec, int child);
+
+void
+ec_shd_index_healer_wake(ec_t *ec);
 
 #endif /* __EC_HEALD_H__ */
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
index 59b4aeefe69..f25760049c3 100644
--- a/xlators/cluster/ec/src/ec.c
+++ b/xlators/cluster/ec/src/ec.c
@@ -322,7 +322,7 @@ ec_get_event_from_state (ec_t *ec)
                 /* If ec is up but some subvolumes are yet to notify, give
                  * grace time for other subvols to notify to prevent start of
                  * I/O which may result in self-heals */
-                if (ec->timer && ec->xl_notify_count < ec->nodes)
+                if (ec->xl_notify_count < ec->nodes)
                         return GF_EVENT_MAXVAL;
 
                 return GF_EVENT_CHILD_UP;
@@ -344,8 +344,8 @@ ec_up (xlator_t *this, ec_t *ec)
         }
 
         ec->up = 1;
-        gf_msg (this->name, GF_LOG_INFO, 0,
-                EC_MSG_EC_UP, "Going UP");
+        gf_msg (this->name, GF_LOG_INFO, 0, EC_MSG_EC_UP, "Going UP");
+
         gf_event (EVENT_EC_MIN_BRICKS_UP, "subvol=%s", this->name);
 }
 
@@ -358,8 +358,8 @@ ec_down (xlator_t *this, ec_t *ec)
         }
 
         ec->up = 0;
-        gf_msg (this->name, GF_LOG_INFO, 0,
-                EC_MSG_EC_DOWN, "Going DOWN");
+        gf_msg (this->name, GF_LOG_INFO, 0, EC_MSG_EC_DOWN, "Going DOWN");
+
         gf_event (EVENT_EC_MIN_BRICKS_NOT_UP, "subvol=%s", this->name);
 }
 
@@ -383,31 +383,38 @@ ec_notify_cbk (void *data)
                 gf_timer_call_cancel (ec->xl->ctx, ec->timer);
                 ec->timer = NULL;
 
+                /* The timeout has expired, so any subvolume that has not
+                 * already reported its state, will be considered to be down.
+                 * We mark as if all bricks had reported. */
+                ec->xl_notify = (1ULL << ec->nodes) - 1ULL;
+                ec->xl_notify_count = ec->nodes;
+
+                /* Since we have marked all subvolumes as notified, it's
+                 * guaranteed that ec_get_event_from_state() will return
+                 * CHILD_UP or CHILD_DOWN, but not MAXVAL. */
                 event = ec_get_event_from_state (ec);
-                /* If event is still MAXVAL then enough subvolumes didn't
-                 * notify, treat it as CHILD_DOWN. */
-                if (event == GF_EVENT_MAXVAL) {
-                        event = GF_EVENT_CHILD_DOWN;
-                        ec->xl_notify = (1ULL << ec->nodes) - 1ULL;
-                        ec->xl_notify_count = ec->nodes;
-                } else if (event == GF_EVENT_CHILD_UP) {
-                        /* Rest of the bricks are still not coming up,
-                         * notify that ec is up. Files/directories will be
-                         * healed as in when they come up. */
+                if (event == GF_EVENT_CHILD_UP) {
+                        /* We are ready to bring the volume up. If there are
+                         * still bricks DOWN, they will be healed when they
+                         * come up. */
                         ec_up (ec->xl, ec);
                 }
 
-                /* CHILD_DOWN should not come here as no grace period is given
-                 * for notifying CHILD_DOWN. */
-
                 propagate = _gf_true;
         }
 unlock:
         UNLOCK(&ec->lock);
 
         if (propagate) {
+                if ((event == GF_EVENT_CHILD_UP) && ec->shd.iamshd) {
+                        /* We have just brought the volume UP, so we trigger
+                         * a self-heal check on the root directory. */
+                        ec_launch_replace_heal (ec);
+                }
+
                 default_notify (ec->xl, event, NULL);
         }
+
 }
 
 void
@@ -442,7 +449,7 @@ ec_pending_fops_completed(ec_t *ec)
         }
 }
 
-static void
+static gf_boolean_t
 ec_set_up_state(ec_t *ec, uintptr_t index_mask, uintptr_t new_state)
 {
         uintptr_t current_state = 0;
@@ -455,7 +462,11 @@ ec_set_up_state(ec_t *ec, uintptr_t index_mask, uintptr_t new_state)
         if (current_state != new_state) {
                 ec->xl_up ^= index_mask;
                 ec->xl_up_count += (current_state ? -1 : 1);
+
+                return _gf_true;
         }
+
+        return _gf_false;
 }
 
 static gf_boolean_t
@@ -498,15 +509,16 @@ ec_upcall(ec_t *ec, struct gf_upcall *upcall)
 int32_t
 ec_notify (xlator_t *this, int32_t event, void *data, void *data2)
 {
-        ec_t             *ec        = this->private;
-        int32_t           idx       = 0;
-        int32_t           error     = 0;
-        glusterfs_event_t old_event = GF_EVENT_MAXVAL;
-        dict_t            *input    = NULL;
-        dict_t            *output   = NULL;
-        gf_boolean_t      propagate = _gf_true;
-        int32_t           orig_event = event;
-        uintptr_t mask = 0;
+        ec_t             *ec              = this->private;
+        int32_t           idx             = 0;
+        int32_t           error           = 0;
+        glusterfs_event_t old_event       = GF_EVENT_MAXVAL;
+        dict_t            *input          = NULL;
+        dict_t            *output         = NULL;
+        gf_boolean_t      propagate       = _gf_true;
+        gf_boolean_t      needs_shd_check = _gf_false;
+        int32_t           orig_event      = event;
+        uintptr_t         mask            = 0;
 
         gf_msg_trace (this->name, 0, "NOTIFY(%d): %p, %p",
                 event, data, data2);
@@ -529,8 +541,6 @@ ec_notify (xlator_t *this, int32_t event, void *data, void *data2)
 
         for (idx = 0; idx < ec->nodes; idx++) {
                 if (ec->xl_list[idx] == data) {
-                        if (event == GF_EVENT_CHILD_UP)
-                                ec_selfheal_childup (ec, idx);
                         break;
                 }
         }
@@ -556,17 +566,27 @@ ec_notify (xlator_t *this, int32_t event, void *data, void *data2)
 
                 mask = 1ULL << idx;
                 if (event == GF_EVENT_CHILD_UP) {
-                    ec_set_up_state(ec, mask, mask);
+                        /* We need to trigger a selfheal if a brick changes
+                         * to UP state. */
+                        needs_shd_check = ec_set_up_state(ec, mask, mask);
                 } else if (event == GF_EVENT_CHILD_DOWN) {
-                    ec_set_up_state(ec, mask, 0);
+                        ec_set_up_state(ec, mask, 0);
                 }
 
                 event = ec_get_event_from_state (ec);
 
-                if (event == GF_EVENT_CHILD_UP && !ec->up) {
-                        ec_up (this, ec);
-                } else if (event == GF_EVENT_CHILD_DOWN && ec->up) {
-                        ec_down (this, ec);
+                if (event == GF_EVENT_CHILD_UP) {
+                        if (!ec->up) {
+                                ec_up (this, ec);
+                        }
+                } else {
+                        /* If the volume is not UP, it's irrelevant if one
+                         * brick has come up. We cannot heal anything. */
+                        needs_shd_check = _gf_false;
+
+                        if ((event == GF_EVENT_CHILD_DOWN) && ec->up) {
+                                ec_down (this, ec);
+                        }
                 }
 
                 if (event != GF_EVENT_MAXVAL) {
@@ -585,14 +605,13 @@ unlock:
 
 done:
         if (propagate) {
+                if (needs_shd_check && ec->shd.iamshd) {
+                        ec_launch_replace_heal (ec);
+                }
+
                 error = default_notify (this, event, data);
         }
 
-        if (ec->shd.iamshd &&
-            ec->xl_notify_count == ec->nodes &&
-            event == GF_EVENT_CHILD_UP) {
-                ec_launch_replace_heal (ec);
-        }
 out:
         return error;
 }