refs #5802, merge probabilistic PAIR clock decrementing from tokudb.fb to main.

git-svn-id: file:///svn/toku/tokudb@51187 c7de825b-a66e-492c-adef-691d508d4ae1

ft/cachetable-internal.h

@@ -9,6 +9,7 @@
 #ident "The technology is licensed by the Massachusetts Institute of Technology, Rutgers State University of New Jersey, and the Research Foundation of State University of New York at Stony Brook under United States of America Serial No. 11/760379 and to the patents and/or patent applications resulting from it."
 
 #include "background_job_manager.h"
+#include <portability/toku_random.h>
 #include <util/frwlock.h>
 #include <util/kibbutz.h>
 #include <util/nb_mutex.h>
@@ -420,6 +421,10 @@ class evictor {
     int64_t m_high_size_watermark; // if cachetable grows to this size, client threads sleep upon adding data
     int64_t m_high_size_hysteresis; // if > cachetable size, then sleeping client threads may wake up
 
+    // used to calculate random numbers
+    struct random_data m_random_data;
+    char m_random_statebuf[64];
+
     // mutex that protects fields listed immedietly below
     toku_mutex_t m_ev_thread_lock;
     // the eviction thread

ft/cachetable.cc

@@ -3484,10 +3484,14 @@ void evictor::init(long _size_limit, pair_list* _pl, KIBBUTZ _kibbutz, uint32_t
     m_ev_thread_is_running = false;
     m_period_in_seconds = eviction_period;
 
+    unsigned int seed = (unsigned int) time(NULL);
+    int r = myinitstate_r(seed, m_random_statebuf, sizeof m_random_statebuf, &m_random_data);
+    assert_zero(r);
+
     // start the background thread    
     m_run_thread = true;
     m_num_eviction_thread_runs = 0;
-    int r = toku_pthread_create(&m_ev_thread, NULL, eviction_thread, this); 
+    r = toku_pthread_create(&m_ev_thread, NULL, eviction_thread, this); 
     assert_zero(r);
 }
 
@@ -3727,6 +3731,7 @@ void evictor::run_eviction(){
 // on exit, the same conditions must apply
 //
 bool evictor::run_eviction_on_pair(PAIR curr_in_clock) {
+    double avg_pair_size;
     bool ret_val = false;
     // function meant to be called on PAIR that is not being accessed right now
     CACHEFILE cf = curr_in_clock->cachefile;
@@ -3742,16 +3747,25 @@ bool evictor::run_eviction_on_pair(PAIR curr_in_clock) {
         bjm_remove_background_job(cf->bjm);
         goto exit;
     }
-    
+
+    avg_pair_size = m_size_current / m_pl->m_n_in_table;
     // now that we have the pair mutex we care about, we can
     // release the read list lock and reacquire it at the end of the function
     m_pl->read_list_unlock();
     ret_val = true;
     if (curr_in_clock->count > 0) {
-        curr_in_clock->count--;
+        if (curr_in_clock->attr.size >= avg_pair_size) {
+            curr_in_clock->count--;
+        } else {
+            int32_t rnd = myrandom_r(&m_random_data);
+            double prob = curr_in_clock->attr.size / avg_pair_size;
+            if ((prob * 10000) < (rnd % 10000)) {
+                curr_in_clock->count--;
+            }
+        }
         // call the partial eviction callback
         curr_in_clock->value_rwlock.write_lock(true);
-    
+
         void *value = curr_in_clock->value_data;
         void* disk_data = curr_in_clock->disk_data;
         void *write_extraargs = curr_in_clock->write_extraargs;

portability/toku_random.h

@@ -35,7 +35,7 @@ myrandom_r(struct random_data *buf)
 struct random_data {
     unsigned short xsubi[3];
 };
-static int
+static inline int
 myinitstate_r(unsigned int seed, char *UU(statebuf), size_t UU(statelen), struct random_data *buf)
 {
     buf->xsubi[0] = (seed & 0xffff0000) >> 16;