From: Pavan Deolasee <pavan.deolasee@gmail.com>
Date: Wed, 6 Jan 2016 13:02:06 +0000 (+0530)
Subject: End global transaction on the GTM before releasing locks.
X-Git-Tag: XL9_5_R1BETA1~114
X-Git-Url: http://git.postgresql.org/gitweb/static/gitweb.js?a=commitdiff_plain;h=b74a34e036dc3d789201c59a0bea7cd82458c7c9;p=postgres-xl.git

End global transaction on the GTM before releasing locks.

Since other backends could be waiting for locks to be released and they must
see end of the transaction, not just locally but also on the GTM (say because
it takes a new snapshot from the GTM, especially for catalog scans). There are
some concerns about doing it the way its done in this patch because we now
exchange messages with the GTM while holding interrupts. But we don't know if
thats really a problem
---

diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c
index 4e0d6daa41..fb15289156 100644
--- a/src/backend/access/transam/xact.c
+++ b/src/backend/access/transam/xact.c
@@ -2545,6 +2545,21 @@ CommitTransaction(void)
 	}
 #endif
 
+	/*
+	 * End the transaction on the GTM before releasing the locks. This would
+	 * ensure that any other backend which might be waiting for our locks,
+	 * would see end of the global transaction before acquiring the locks.
+	 *
+	 * XXX Earlier we used to do this at the very end of this function. But
+	 * that leads to various issues (such as "tuple concurrently updated"
+	 * errors in concurrent ANALYZE). It seems like a good idea to end the
+	 * global transaction before releasing the locks from correctness
+	 * perspective. But we are now doing network calls while holding
+	 * interrupts. That can lead to some unpleasant hangs. So lets be wary
+	 * about this possibility
+	 */
+	AtEOXact_GlobalTxn(true);
+
 	ResourceOwnerRelease(TopTransactionResourceOwner,
 						 RESOURCE_RELEASE_LOCKS,
 						 true, true);
@@ -2624,14 +2639,9 @@ CommitTransaction(void)
 
 	RESUME_INTERRUPTS();
 
-	/*
-	 * XXX We now close the main and auxilliary transaction (if any) on the
-	 * GTM. We do this after resuming interrupts to ensure that we don't end
-	 * blocking forever on the communication channel. But we need to see if
-	 * this is safe in all cases (TODO)
-	 */
-	AtEOXact_GlobalTxn(true);
+#ifdef PGXC
 	AtEOXact_Remote();
+#endif
 }
 
 /*
@@ -3262,6 +3272,12 @@ AbortTransaction(void)
 		AtEOXact_RelationCache(false);
 		AtEOXact_Inval(false);
 		AtEOXact_MultiXact();
+
+		/* See comments in CommitTransaction */
+#ifdef XCP
+		AtEOXact_GlobalTxn(false);
+#endif
+
 		ResourceOwnerRelease(TopTransactionResourceOwner,
 							 RESOURCE_RELEASE_LOCKS,
 							 false, true);
@@ -3292,7 +3308,6 @@ AbortTransaction(void)
 	RESUME_INTERRUPTS();
 
 #ifdef PGXC
-	AtEOXact_GlobalTxn(false);
 	AtEOXact_Remote();
 #endif