[Maria-developers] MariaDB 5.5.35
Hi, I started to notice this on my error log since i upgraded from 5.5.34 to 5.5.35: 140203 5:26:25 InnoDB: Failed to set O_DIRECT on file /dev/shm/#sql7e8f6_1a94bc_10.ibd: CREATE: Invalid argument, continuing anyway 140203 5:26:25 InnoDB: O_DIRECT is known to result in 'Invalid argument' on Linux on tmpfs, see MySQL Bug#26662 I never seen this error/warning until 5.5.35. Should this be normal ? Thank you in advance. Alex
Am 03.02.2014 18:08, schrieb ImperialNetwork:
I started to notice this on my error log since i upgraded from 5.5.34 to 5.5.35:
140203 5:26:25 InnoDB: Failed to set O_DIRECT on file /dev/shm/#sql7e8f6_1a94bc_10.ibd: CREATE: Invalid argument, continuing anyway 140203 5:26:25 InnoDB: O_DIRECT is known to result in 'Invalid argument' on Linux on tmpfs, see MySQL Bug#26662
I never seen this error/warning until 5.5.35. Should this be normal?
output of /usr/bin/df -hT would be a good idea if you heave /tmp as tmpfs like more and more linux distributions set "tmpdir" in my.cnf somewhere else, O_DIRECT can not work on tmpfs by definition and that there was before no warning don't mean that it ever worked before there where many warnings in the past 6 months about config options which are deprecated or not work as someone may think the difference is "works not" silent or not silent in most cases
Hi, Using a tmpfs directory as the tmpdir has its interest, changing it only because of warnings issued for temporaries InnoDB tables doesn't seems to be a "solution". Anyway there is no point on using O_DIRECT on a filesystem in ram (on its "normal" usage at least). Issuing the warning once when the server starts and check the tmpdir could be enough. Regards. On 03/02/2014 19:09, Reindl Harald wrote:
Am 03.02.2014 18:08, schrieb ImperialNetwork:
I started to notice this on my error log since i upgraded from 5.5.34 to 5.5.35:
140203 5:26:25 InnoDB: Failed to set O_DIRECT on file /dev/shm/#sql7e8f6_1a94bc_10.ibd: CREATE: Invalid argument, continuing anyway 140203 5:26:25 InnoDB: O_DIRECT is known to result in 'Invalid argument' on Linux on tmpfs, see MySQL Bug#26662
I never seen this error/warning until 5.5.35. Should this be normal?
output of /usr/bin/df -hT would be a good idea
if you heave /tmp as tmpfs like more and more linux distributions set "tmpdir" in my.cnf somewhere else, O_DIRECT can not work on tmpfs by definition and that there was before no warning don't mean that it ever worked before
there where many warnings in the past 6 months about config options which are deprecated or not work as someone may think
the difference is "works not" silent or not silent in most cases
Am 04.02.2014 10:01, schrieb Jean Weisbuch:
Using a tmpfs directory as the tmpdir has its interest
especially if it leads to swap or is not large enough for a temp-table for a random operation leading to errors keep in mind that innodb_buffer_pool and other processes are fighting for the same RAM
changing it only because of warnings issued for temporaries InnoDB tables doesn't seems to be a "solution"
it's the solution for exactly the mentioned problem the other one is not using O_DIRECT O_DIRECT don't work on tmpfs by definition
On 03/02/2014 19:09, Reindl Harald wrote:
Am 03.02.2014 18:08, schrieb ImperialNetwork:
I started to notice this on my error log since i upgraded from 5.5.34 to 5.5.35:
140203 5:26:25 InnoDB: Failed to set O_DIRECT on file /dev/shm/#sql7e8f6_1a94bc_10.ibd: CREATE: Invalid argument, continuing anyway 140203 5:26:25 InnoDB: O_DIRECT is known to result in 'Invalid argument' on Linux on tmpfs, see MySQL Bug#26662
I never seen this error/warning until 5.5.35. Should this be normal?
output of /usr/bin/df -hT would be a good idea
if you heave /tmp as tmpfs like more and more linux distributions set "tmpdir" in my.cnf somewhere else, O_DIRECT can not work on tmpfs by definition and that there was before no warning don't mean that it ever worked before
there where many warnings in the past 6 months about config options which are deprecated or not work as someone may think
the difference is "works not" silent or not silent in most cases
participants (3)
-
ImperialNetwork -
Jean Weisbuch -
Reindl Harald