Subject: Re: fsync error handling in pg_receivewal, pg_recvlogical
To: Michael Paquier <michael@paquier.xyz>
References: <9b49fe44-8f3e-eca9-5914-29e9e99030bf@2ndquadrant.com>
 <20190329130505.GG1954@paquier.xyz>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>
From: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Openpgp: preference=signencrypt
Autocrypt: addr=peter.eisentraut@2ndquadrant.com; prefer-encrypt=mutual;
 keydata=
 xsFNBFUUt6kBEADZd9NNTBJbHmAXXJgPGOdd6Q73l80oifIVgn2tRxP9dYkCfTlo/yLqbKTl
 JNqTcyM6XrJDnB2sMEUGYCewYZXl+9C4bBJWyNqcEI1ae9u71vEq/fknFLArJ+2wN+f7QuRx
 Q8d94ZJAiJQ5g3fNk2HVeedCQCKP5dCzU+ZLt2dbsx1PDZLjTO9gf+JswWuGFb5wY5PZVM5V
 hq66Xk7ZJzHuARacza2RT7J2fkMmQBY+kr7L98Perb0GhafnlhXB0PvUB+N3xpmggMfwuhTu
 BngS/G/r3GvemxJuFxouw8xsMqrsfdksJXA2gLXYtwRCssdPV3PO0rlD9twqDo8URfIJ5fb6
 1g/MV8aTOKjg3Q8MpJPtkwJqN4o3PvbwYdedeFnOVkkYAJdu/aX1v1BeCBDu7Hy3sF52scD3
 pQbyQSl9vUDpTLarASAxNbAV0NRJE8dp4vfW6AH1ccryGiFqw/czPK/iN0+MT8wsjAW6WW4w
 hvjm5UPA1KTwi3ZT8eT5uycuH7QpuNk7LnGGXQ6KaRALFywzkr0A0V0AvFNa+l4wstMaXaB6
 dmMXyF0/hfryqyOeu1gAdq4q0/PNC2dPgb/6uoh4eMjN6A5GXcprilhz5DulmsilY21hfSoU
 S64ODed96pKhprsOxjZiDRsN1QM5MwB+uwVWUNfsX7fIxE0BrwARAQABzSdQZXRlciBFaXNl
 bnRyYXV0IDxwZXRlckBlaXNlbnRyYXV0Lm9yZz7CwXoEEwEIACQCGwMFCwkIBwMFFQoJCAsF
 FgIDAQACHgECF4AFAlUUuKQCGQEACgkQoZ8SQ+xBJp8llw//RbGoXnF72FZTlOFsnNyxnWbF
 vViPCfdAlZl1GGK9OYCvV1dp/956HuhB0wpaGQHiBmQfOPJf/jLOPjXFYm2iEChEycrBChuE
 j4TSYemMp5BSPILYk9Uwd16kVD/+U6fQfK6H5pOhfBvbM6asMnmpY2wQuvktxHvaGlytFw84
 EzFWNZtAZfxkw28w3AhIQkKe9hJanfyDL750RcbyEr8AAHty5IErrvS7tM5vQwIeh5BF26ND
 CyKPRdZgOgMpcwFfROKBDpAt052CgWIpNdhBdfIKlkCs4/ILxmC3rT/Mb/qZiHcoMMwEu6k+
 T7vRAFZZAk62Yrr/U8/vUGKlzRZQoo86fyojCfI0ibCIc0z8rwCVY9P40G9wD9BTCZS/z8Vh
 jjpBlNnIKw2hM6hbxavKndCJjFFJD08rXrTyrLF/of18xD37vNdYooP9VATB5QjOWFRrscSZ
 4a89e+to3m7YVnd4xdXwGp47dYusoax84C8tg1gZrH38ol9dPmnKX6GZrXeS7tH9k8VeUpJ8
 HQbRThQRWAT936QpYbLConBmOtmbLmjQOGtiXPo1oK57X/7U/WBja5vWmm2q+Yxtt+3uHGVg
 DdDU4CJpXo4lX0v2n3QV5EBSW9hjdF3bNfl6GFBb1pPEhqNcMDUc7zkx/2uENdAdk/BnqMgH
 umip7b0pVfvOwU0EVRS3qQEQAJf6p0P7D+hMYUgCPkvgrcz6OJYsAYdUMUmokfqmkKwK0oJM
 ew8TNGM53ocKAgQGma5f10rwbZNPlpVPNRX5hwUmwVlcwSut0oE+JfUBH7mpysWvpF05RKIy
 q9RCoJZrAbXS7Tig56A1QD8IcM00rERZMcCOlI1+q9m4u9AOI+0LyGGSH+VvoyDHD/Jew4R2
 TmPEF8Pu7lYiNYK88Q6u0kBkc+Tz3BNlyc6fi3HxnFfp0b75HLXPSTu2SwA/0Q31wHZXwrEU
 BR9KgSn/qbgy8ydPc97i82fhiFE3cbGexfPrQSMYmDGi+dcQ0s+AMQwC3XyMC0JkHpYUuFkR
 QdlIvaIhgFUUN2Vr4xF2ADaNTNFc4DYaiPHFAgHZw+3pZWqS4uIiWgJynrEaHw4F3QWU8/JB
 iyEWDhNq4WBrRA0u914TCCAoFOOKMnRdl8eJrRQKM/5/lXbHFjjlI/LDKhLRFsuQsnJJ3331
 RjCzpHPZG5/p5KR1RHiAXa0p/DW5lXyq21hS3kjDlleoHqOxCN/cy0Jgz7qLDq1vZoCl67Sy
 RAH60c2Xu4may9SVltmot0OzWVHdrfS+85FQGp9580DaeLio7YMFKy2bfEvwQJ/c9xD2QPB+
 oSb3V8ueSznXBG5+dyfOsZFRPxceA0wA2rSWZSjxLdKxvyGTj9vsCjaDu13rABEBAAHCwV8E
 GAEIAAkFAlUUt6kCGwwACgkQoZ8SQ+xBJp+HJhAAxmSTsC5y7p9Zu+nnLpucv/h88oEx985e
 qWeXzQiwM3pfRZS+PdTH9TyzHK0KOgKoAiubk0vt+BrT91D9d0G5kVQaTctAErGBoMkTaZJT
 XVEcNphiPzysdSqvprjojMp1y7rG8AP+Xd6m0rE1LRnkRRm8HyI8vdXsuLQPITGWXDLMjaNq
 AOV3BUnzkgtjMuIYw7SCr+12ZOb5dEKvaG4kjC4kaJ/TLbuBhVANcXkC90yp0AOZLs/cB1Dj
 OAXno/RUUruJ3r27a+pFv768UAi9QWcdw58yPxreMNMLqAgsS9oTHwHfGNWcB2dJ4UW+GMLJ
 hTf9TjQ2f+N/sIx5XMQ9pYw6CAJMK29qPx6/ZayrPxf22EXf0D3zTwIEZgwfSkynFVwV25L7
 awQq60fifdMYheSxq7GbhpsixmcD+e161cCmLIt75YDt5bQ0B1yf7EmxG8n7pGZAOM6LoNsU
 fkSldba5XVwh8LuN5mK0nQBELGW6Os4TDq4REggLpfcvFC215zpuo0iCGnmE/Bns4nowqUSP
 u4tkWL+nxW4qgv8wo8HJ52jd79EJ8iqfIBoidBfy9AIUiEUApGQzWPu6yWcI8fY0rN5D+tc+
 Yw1Zt6Mm+huYJpgCRsTcfzEcahnBn521CWcWvhW1ff3KFileFMtNmnBBhMncmRKHphtTKEpN
 3jI=
Organization: 2ndQuadrant
Message-ID: <43bba31e-c520-2de0-1b02-5672bc7faf70@2ndquadrant.com>
Date: Tue, 25 Jun 2019 14:23:05 +0200
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.14; rv:52.0)
 Gecko/20100101 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <20190329130505.GG1954@paquier.xyz>
Content-Type: multipart/mixed;
 boundary="------------FD58F2D0F6ABC7B4E776A847"
Content-Language: en-US
Precedence: bulk

This is a multi-part message in MIME format.
--------------FD58F2D0F6ABC7B4E776A847
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit

On 2019-03-29 14:05, Michael Paquier wrote:
> Yes, I think that we are going to need an equivalent of that for all
> frontend tools.  At various degrees, making sure that a fsync happens
> is also important for pg_dump, pg_basebackup, pg_rewind and
> pg_checksums so it is not only a problem of the two tools you mention.
> It seems to me that the most correct way to have those failures would
> be to use directly exit(EXIT_FAILURE) in file_utils.c where
> appropriate.

Yeah, there is more to do.  The reason I'm focusing on these two right
now is that they would typically run as a background service, and a
clean exit is most important there.  In the other cases, the program
runs more often in the foreground and you can see error messages.  There
are also some cases where fsync() failures are intentionally ignored
((void) casts), so some of that would need to be investigated further.

Here is a patch to get started.  Note that these calls don't go through
file_utils.c, so it's a separate issue anyway.

-- 
Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

--------------FD58F2D0F6ABC7B4E776A847
Content-Type: text/plain; charset=UTF-8; x-mac-type="0"; x-mac-creator="0";
 name="0001-Handle-fsync-failures-in-pg_receivewal-and-pg_recvlo.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
 filename*0="0001-Handle-fsync-failures-in-pg_receivewal-and-pg_recvlo.pa";
 filename*1="tch"

From 8f275321775f41d4a46ea6e1b10765c68fe2653d Mon Sep 17 00:00:00 2001
From: Peter Eisentraut <peter@eisentraut.org>
Date: Tue, 25 Jun 2019 14:19:26 +0200
Subject: [PATCH] Handle fsync failures in pg_receivewal and pg_recvlogical

It is not safe to simply report an fsync error and continue.  We must
exit the program instead.
---
 src/bin/pg_basebackup/pg_recvlogical.c |  4 ++--
 src/bin/pg_basebackup/receivelog.c     | 12 ++++++------
 src/bin/pg_basebackup/walmethods.c     |  2 +-
 3 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/src/bin/pg_basebackup/pg_recvlogical.c b/src/bin/pg_basebackup/pg_recvlogical.c
index b029118bf6..9d8f8c74f8 100644
--- a/src/bin/pg_basebackup/pg_recvlogical.c
+++ b/src/bin/pg_basebackup/pg_recvlogical.c
@@ -192,8 +192,8 @@ OutputFsync(TimestampTz now)
 
 	if (fsync(outfd) != 0)
 	{
-		pg_log_error("could not fsync file \"%s\": %m", outfile);
-		return false;
+		pg_log_fatal("could not fsync file \"%s\": %m", outfile);
+		exit(2);
 	}
 
 	return true;
diff --git a/src/bin/pg_basebackup/receivelog.c b/src/bin/pg_basebackup/receivelog.c
index 2fd16421aa..81fc1b0a01 100644
--- a/src/bin/pg_basebackup/receivelog.c
+++ b/src/bin/pg_basebackup/receivelog.c
@@ -134,10 +134,10 @@ open_walfile(StreamCtl *stream, XLogRecPtr startpoint)
 			/* fsync file in case of a previous crash */
 			if (stream->walmethod->sync(f) != 0)
 			{
-				pg_log_error("could not fsync existing write-ahead log file \"%s\": %s",
+				pg_log_fatal("could not fsync existing write-ahead log file \"%s\": %s",
 							 fn, stream->walmethod->getlasterror());
 				stream->walmethod->close(f, CLOSE_UNLINK);
-				return false;
+				exit(2);
 			}
 
 			walfile = f;
@@ -763,9 +763,9 @@ HandleCopyStream(PGconn *conn, StreamCtl *stream,
 		{
 			if (stream->walmethod->sync(walfile) != 0)
 			{
-				pg_log_error("could not fsync file \"%s\": %s",
+				pg_log_fatal("could not fsync file \"%s\": %s",
 							 current_walfile_name, stream->walmethod->getlasterror());
-				goto error;
+				exit(2);
 			}
 			lastFlushPosition = blockpos;
 
@@ -1015,9 +1015,9 @@ ProcessKeepaliveMsg(PGconn *conn, StreamCtl *stream, char *copybuf, int len,
 			 */
 			if (stream->walmethod->sync(walfile) != 0)
 			{
-				pg_log_error("could not fsync file \"%s\": %s",
+				pg_log_fatal("could not fsync file \"%s\": %s",
 							 current_walfile_name, stream->walmethod->getlasterror());
-				return false;
+				exit(2);
 			}
 			lastFlushPosition = blockpos;
 		}
diff --git a/src/bin/pg_basebackup/walmethods.c b/src/bin/pg_basebackup/walmethods.c
index 83b520898b..d8abea6dc6 100644
--- a/src/bin/pg_basebackup/walmethods.c
+++ b/src/bin/pg_basebackup/walmethods.c
@@ -864,7 +864,7 @@ tar_close(Walfile f, WalCloseMethod method)
 
 	/* Always fsync on close, so the padding gets fsynced */
 	if (tar_sync(f) < 0)
-		return -1;
+		exit(2);
 
 	/* Clean up and done */
 	pg_free(tf->pathname);
-- 
2.22.0


--------------FD58F2D0F6ABC7B4E776A847--