MB-34955: Reduce size of ADM's CompletedQueue

The internal folly::USPSCQueue defaults to create a single
segment, the segment holds 256 entries (256 SyncWrites)

This default setting impacts the bucket overhead, for a
single node an empty bucket (i.e. 1024 vbuckets) the
introduction of the folly::USPSCQueue increased mem_used
by 24MB, that increase was first seen from 6.5.0-3621 (22MB)
to 6.5.0-3622 (46MB).

This commit sets the initial size of the queue to store 32
entries per segment, bringing the bucket mem_used down to
approx 23MB

Change-Id: I08a0be01db0d94d6efcbb7f2e449e269b12ed440
Reviewed-on: http://review.couchbase.org/113375
Reviewed-by: Dave Rigby <[email protected]>
Tested-by: Build Bot <[email protected]>

Author: jimwwalker

engines/ep/src/durability/active_durability_monitor.cc

@@ -445,8 +445,8 @@ class ActiveDurabilityMonitor::ResolvedQueue {
 
 private:
     // Unbounded, Single-producer, single-consumer Queue of SyncWrite objects,
-    // non-blocking variant.
-    using Queue = folly::USPSCQueue<DurabilityMonitor::SyncWrite, false>;
+    // non-blocking variant. Initially holds 2^5 (32) SyncWrites
+    using Queue = folly::USPSCQueue<DurabilityMonitor::SyncWrite, false, 5>;
     Queue queue;
     // Track the highest Enqueued Seqno to enforce enqueue ordering.
     Monotonic<int64_t> highEnqueuedSeqno = {0};