pg_repack.git / summary / log / commit / refs

commit    ef4dc5a1fbb25eedd8991fbbfaf4b70e8d33b5cd
Author:   Josh Kupershmidt <schmiddy@gmail.com>
Date:     Mon Jan 12 01:59:13 2015 +0000

    Tweak apply_log() loop so that we don't wait until getting
    all the way down to 0 rows processed before performing the
    table swap step.


 bin/pg_repack.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/bin/pg_repack.c b/bin/pg_repack.c
index d2cc161..4095037 100644
--- a/bin/pg_repack.c
+++ b/bin/pg_repack.c
@@ -48,6 +48,11 @@ const char *PROGRAM_VERSION = "unknown";
  */
 #define APPLY_COUNT		1000
 
+/* Once we get down to seeing fewer than this many tuples in the
+ * log table, we'll say that we're ready to perform the switch.
+ */
+#define MIN_TUPLES_BEFORE_SWITCH	20
+
 /* poll() or select() timeout, in seconds */
 #define POLL_TIMEOUT    3
 
@@ -1305,7 +1310,16 @@ repack_one_table(repack_table *table, const char *orderby)
 	for (;;)
 	{
 		num = apply_log(connection, table, APPLY_COUNT);
-		if (num > 0)
+
+		/* We'll keep applying tuples from the log table in batches
+		 * of APPLY_COUNT, until applying a batch of tuples
+		 * (via LIMIT) results in our having applied
+		 * MIN_TUPLES_BEFORE_SWITCH or fewer tuples. We don't want to
+		 * get stuck repetitively applying some small number of tuples 
+		 * from the log table as inserts/updates/deletes may be
+		 * constantly coming into the original table.
+		 */
+		if (num > MIN_TUPLES_BEFORE_SWITCH)
 			continue;	/* there might be still some tuples, repeat. */
 
 		/* old transactions still alive ? */


[parent: 8f61e44efd34]