pgsql: Use XLOG_BLCKSZ in pg_test_fsync, rather than our own define, bu
Use XLOG_BLCKSZ in pg_test_fsync, rather than our own define, but verify
it is 8k as expected.
Branch
------
master
Details
-------
http://git.postgresql.org/gitweb?p=postgresql.git;a=commitdiff;h=9dd7933937a076ce7573944b8d1d42e618163440
Modified Files
--------------
contrib/pg_test_fsync/pg_test_fsync.c | 24 +++++++++++++-----------
1 files changed, 13 insertions(+), 11 deletions(-)
Bruce Momjian <bruce@momjian.us> writes:
Use XLOG_BLCKSZ in pg_test_fsync, rather than our own define, but verify
it is 8k as expected.
-#define WRITE_SIZE (8 * 1024) /* 8k */
+#if XLOG_BLCKSZ != 8 * 1024 /* 8k */
+#error Unknown block size for test.
+#endifThis seems like a pretty awful idea. Aren't you aware that XLOG_BLCKSZ
is settable from a configure option? You just broke the ability to
build the tree with a non-default configuration setting.
If you aren't willing to deal with a variable value for the block size,
please revert this patch.
regards, tom lane
Tom Lane wrote:
Bruce Momjian <bruce@momjian.us> writes:
Use XLOG_BLCKSZ in pg_test_fsync, rather than our own define, but verify
it is 8k as expected.-#define WRITE_SIZE (8 * 1024) /* 8k */ +#if XLOG_BLCKSZ != 8 * 1024 /* 8k */ +#error Unknown block size for test. +#endifThis seems like a pretty awful idea. Aren't you aware that XLOG_BLCKSZ
is settable from a configure option? You just broke the ability to
build the tree with a non-default configuration setting.If you aren't willing to deal with a variable value for the block size,
please revert this patch.
The problem is that I have hard-coded 8k into various text strings and I
didn't want to make that variable. How should it behave if they are
using a non-8k wal buffer size? Should it still use 8k or not? I
figured throwing an error would at least alert them to the mismatch.
Good point on the build problem --- I had not thought of that. Throwing
an error when running makes more sense, but let's figure out what it
should do.
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +
Bruce Momjian <bruce@momjian.us> writes:
Tom Lane wrote:
If you aren't willing to deal with a variable value for the block size,
please revert this patch.
The problem is that I have hard-coded 8k into various text strings and I
didn't want to make that variable. How should it behave if they are
using a non-8k wal buffer size? Should it still use 8k or not? I
figured throwing an error would at least alert them to the mismatch.
Well, as I said, if you aren't willing to put effort into that point,
just revert the patch. Making the program refuse to do anything doesn't
help *anyone*. Stats taken using a fixed 8K blocksize are better than
no stats at all.
regards, tom lane
Tom Lane wrote:
Bruce Momjian <bruce@momjian.us> writes:
Tom Lane wrote:
If you aren't willing to deal with a variable value for the block size,
please revert this patch.The problem is that I have hard-coded 8k into various text strings and I
didn't want to make that variable. How should it behave if they are
using a non-8k wal buffer size? Should it still use 8k or not? I
figured throwing an error would at least alert them to the mismatch.Well, as I said, if you aren't willing to put effort into that point,
just revert the patch. Making the program refuse to do anything doesn't
help *anyone*. Stats taken using a fixed 8K blocksize are better than
no stats at all.
Sure I am willing to fix it. Should I have it always use the value of
XLOG_BLCKSZ for its tests, and adjust the output text accordingly?
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +
Bruce Momjian <bruce@momjian.us> writes:
Sure I am willing to fix it. Should I have it always use the value of
XLOG_BLCKSZ for its tests, and adjust the output text accordingly?
Makes sense to me.
regards, tom lane
Tom Lane wrote:
Bruce Momjian <bruce@momjian.us> writes:
Sure I am willing to fix it. Should I have it always use the value of
XLOG_BLCKSZ for its tests, and adjust the output text accordingly?Makes sense to me.
OK, done with attached patch. I also cleaned up the open_sync size test
output. The patch was easy. Thanks for the ideas.
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +