From 1c94e12c3d7615e4859eddde88ed7e53cd127d32 Mon Sep 17 00:00:00 2001
From: dormando <dormando@rydia.net>
Date: Sat, 18 Aug 2012 12:54:01 -0700
Subject: item locks now lock hash table buckets

expansion requires switching to a global lock temporarily, so all buckets have
a covered read lock.

slab rebalancer is paused during hash table expansion.

internal item "trylocks" are always issued, and tracked as the hash power
variable can change out from under it.
---
 assoc.c | 30 +++++++++++++++++++++++++-----
 1 file changed, 25 insertions(+), 5 deletions(-)

(limited to 'assoc.c')

diff --git a/assoc.c b/assoc.c
index bcaf6f2..0edce2b 100644
--- a/assoc.c
+++ b/assoc.c
@@ -32,7 +32,7 @@ typedef  unsigned long  int  ub4;   /* unsigned 4-byte quantities */
 typedef  unsigned       char ub1;   /* unsigned 1-byte quantities */
 
 /* how many powers of 2's worth of buckets we use */
-static unsigned int hashpower = HASHPOWER_DEFAULT;
+unsigned int hashpower = HASHPOWER_DEFAULT;
 
 #define hashsize(n) ((ub4)1<<(n))
 #define hashmask(n) (hashsize(n)-1)
@@ -51,6 +51,7 @@ static unsigned int hash_items = 0;
 
 /* Flag: Are we in the middle of expanding now? */
 static bool expanding = false;
+static bool started_expanding = false;
 
 /*
  * During expansion we migrate values with bucket granularity; this is how
@@ -136,13 +137,19 @@ static void assoc_expand(void) {
         stats.hash_bytes += hashsize(hashpower) * sizeof(void *);
         stats.hash_is_expanding = 1;
         STATS_UNLOCK();
-        pthread_cond_signal(&maintenance_cond);
     } else {
         primary_hashtable = old_hashtable;
         /* Bad news, but we can keep running. */
     }
 }
 
+static void assoc_start_expand(void) {
+    if (started_expanding)
+        return;
+    started_expanding = true;
+    pthread_cond_signal(&maintenance_cond);
+}
+
 /* Note: this isn't an assoc_update.  The key must not already exist to call this */
 int assoc_insert(item *it, const uint32_t hv) {
     unsigned int oldbucket;
@@ -161,7 +168,7 @@ int assoc_insert(item *it, const uint32_t hv) {
 
     hash_items++;
     if (! expanding && hash_items > (hashsize(hashpower) * 3) / 2) {
-        assoc_expand();
+        assoc_start_expand();
     }
 
     MEMCACHED_ASSOC_INSERT(ITEM_key(it), it->nkey, hash_items);
@@ -201,6 +208,7 @@ static void *assoc_maintenance_thread(void *arg) {
 
         /* Lock the cache, and bulk move multiple buckets to the new
          * hash table. */
+        item_lock_global();
         mutex_lock(&cache_lock);
 
         for (ii = 0; ii < hash_bulk_move && expanding; ++ii) {
@@ -230,12 +238,24 @@ static void *assoc_maintenance_thread(void *arg) {
             }
         }
 
+        mutex_unlock(&cache_lock);
+        item_unlock_global();
+
         if (!expanding) {
+            /* finished expanding. tell all threads to use fine-grained locks */
+            switch_item_lock_type(ITEM_LOCK_GRANULAR);
+            started_expanding = false;
+            slabs_rebalancer_resume();
             /* We are done expanding.. just wait for next invocation */
             pthread_cond_wait(&maintenance_cond, &cache_lock);
+            /* Before doing anything, tell threads to use a global lock */
+            mutex_unlock(&cache_lock);
+            slabs_rebalancer_pause();
+            switch_item_lock_type(ITEM_LOCK_GLOBAL);
+            mutex_lock(&cache_lock);
+            assoc_expand();
+            mutex_unlock(&cache_lock);
         }
-
-        mutex_unlock(&cache_lock);
     }
     return NULL;
 }
-- 
cgit v1.2.1