[bug fix] pg_rewind creates corrupt WAL files, and the standby cannot catch up the primary

Hello,

Our customer hit another bug of pg_rewind with PG 9.5.  The attached patch fixes this.


PROBLEM
========================================

After a long run of successful pg_rewind, the synchronized standby could not catch up the primary forever, emitting the following message repeatedly:

LOG:  XX000: could not read from log segment 000000060000028A00000031, offset 16384: No error


CAUSE
========================================

If the primary removes WAL files that pg_rewind is going to get, pg_rewind leaves 0-byte WAL files in the target directory here:

[libpq_fetch.c]
			case FILE_ACTION_COPY:
				/* Truncate the old file out of the way, if any */
				open_target_file(entry->path, true);
				fetch_file_range(entry->path, 0, entry->newsize);
				break;

pg_rewind completes successfully, create recovery.conf, and then start the standby in the target cluster.  walreceiver receives WAL records and write them to the 0-byte WAL files.  Finally, xlog reader complains that he cannot read a WAL page.


FIX
========================================

pg_rewind deletes the file when it finds that the primary has deleted it.


OTHER THOUGHTS
========================================

BTW, should pg_rewind really copy WAL files from the primary?  If the sole purpose of pg_rewind is to recover an instance to use as a standby, can pg_rewind just remove all WAL files in the target directory, because the standby can get WAL files from the primary and/or archive?

Related to this, shouldn't pg_rewind avoid copying more files and directories like pg_basebackup?  Currently, pg_rewind doesn't copy postmaster.pid, postmaster.opts, and temporary files/directories (pg_sql_tmp/).

Regards
Takayuki Tsunakawa