drupal.org caching too aggressive?
Recently I've posted comments on Drupal.org, and after pressing the submit button I've seen the original post, without a message and without my comment. If I refresh, the comment is there. But at least once I've gone back and resubmitted, because as a user it looks like my post was simply ignored. I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on? Anyone else notice this? -Dave
On Dec 5, 2007, at 11:26 AM, Dave Cohen wrote:
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on?
No. You're probably seeing the page as one of the secondary database servers thinks it exists. There's replication delay, now that we have a cluster of DB servers, not just a single box. One of the costs of scalability...
Anyone else notice this?
Yes. -Derek (dww)
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Derek Wright schrieb:
On Dec 5, 2007, at 11:26 AM, Dave Cohen wrote:
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on?
No. You're probably seeing the page as one of the secondary database servers thinks it exists. There's replication delay, now that we have a cluster of DB servers, not just a single box. One of the costs of scalability...
I don't think this really applies. The only queries that get directed to the slave are pager and search queries. The query that gets comments is neither. Cheers, Gerhard -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFHV0Yffg6TFvELooQRAn22AJ9M1q5ybK1w7FUulg/OJUmGdsC1hQCfU5S6 4Nqq6UqNVYX/PJVZVPn95ys= =0x5m -----END PGP SIGNATURE-----
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Gerhard Killesreiter schrieb:
Derek Wright schrieb:
On Dec 5, 2007, at 11:26 AM, Dave Cohen wrote:
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on? No. You're probably seeing the page as one of the secondary database servers thinks it exists. There's replication delay, now that we have a cluster of DB servers, not just a single box. One of the costs of scalability...
I don't think this really applies. The only queries that get directed to the slave are pager and search queries. The query that gets comments is neither.
That's actually wrong, it is a pager query. We probably shoudl discuss on how to pass a "this query is currently not pager safe"-parameter to the database layer. Cheers, Gerhard -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFHV1OEfg6TFvELooQRAl7RAJ0fh5v2UmFrD36JXinb3LjWYHbpSACePZ4n l2c4k2vvnz31YoAn7Ed61NU= =Dyq/ -----END PGP SIGNATURE-----
I'll be looking at this, but I really doubt its replication delay at this point...especially the stories about 6 minute delays, we only fall even close to that far behind when someone is doing large db imports and I usually yank the slave out of rotation when I get paged for that. However, it is possible so I'll take a look and see if I can setup warnings on a finer grained basis. On Dec 5, 2007 5:42 PM, Gerhard Killesreiter <gerhard@killesreiter.de> wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Gerhard Killesreiter schrieb:
Derek Wright schrieb:
On Dec 5, 2007, at 11:26 AM, Dave Cohen wrote:
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on? No. You're probably seeing the page as one of the secondary database servers thinks it exists. There's replication delay, now that we have a cluster of DB servers, not just a single box. One of the costs of scalability...
I don't think this really applies. The only queries that get directed to the slave are pager and search queries. The query that gets comments is neither.
That's actually wrong, it is a pager query. We probably shoudl discuss on how to pass a "this query is currently not pager safe"-parameter to the database layer.
Cheers, Gerhard
-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux)
iD8DBQFHV1OEfg6TFvELooQRAl7RAJ0fh5v2UmFrD36JXinb3LjWYHbpSACePZ4n l2c4k2vvnz31YoAn7Ed61NU= =Dyq/ -----END PGP SIGNATURE-----
-- Narayan Newton Database Administrator OSU Open Source Lab
In case it helps, when this happens I never see the confirmation message which comes from drupal_set_message(). Which indicates something is happening to $_SESSION data. Usually, when a drupal_set_message() is not seen right away, it will appear eventually (next time a page is viewed with that session still active). -Dave On Wed, 5 Dec 2007 19:40:50 -0800, "Narayan Newton" <nnewton@osuosl.org> said:
I'll be looking at this, but I really doubt its replication delay at this point...
I know this looks not related but perhaps worth your attention, especially in case d.o. uses APC (I wouldn't know). http://drupal.org/node/196810 Basically, behaviour same as described by Dave but noticed at editing nodes, also without Drupal's confirmation msg. Tomas David Cohen wrote:
In case it helps, when this happens I never see the confirmation message which comes from drupal_set_message(). Which indicates something is happening to $_SESSION data. Usually, when a drupal_set_message() is not seen right away, it will appear eventually (next time a page is viewed with that session still active).
-Dave
On Wed, 5 Dec 2007 19:40:50 -0800, "Narayan Newton" <nnewton@osuosl.org> said:
I'll be looking at this, but I really doubt its replication delay at this point...
The APC issue doesn't apply here. It is a problem with the apc_cache module which is still alpha code and not ready for use (would be nice if someone would get interested in it and start working on it!) -Robert Tomas J. Fulopp - Vacilando.org wrote:
I know this looks not related but perhaps worth your attention, especially in case d.o. uses APC (I wouldn't know).
Basically, behaviour same as described by Dave but noticed at editing nodes, also without Drupal's confirmation msg.
Tomas
David Cohen wrote:
In case it helps, when this happens I never see the confirmation message which comes from drupal_set_message(). Which indicates something is happening to $_SESSION data. Usually, when a drupal_set_message() is not seen right away, it will appear eventually (next time a page is viewed with that session still active).
-Dave
On Wed, 5 Dec 2007 19:40:50 -0800, "Narayan Newton" <nnewton@osuosl.org> said:
I'll be looking at this, but I really doubt its replication delay at this point...
One thing I noticed with APC cache is that APC alone speeds up Drupal considerably, but activating the APC module by enabling caching to APC is much slower than standard caching. ----- Original Message ----- From: "Robert Douglass" <rob@robshouse.net> To: <development@drupal.org>; <tomi@vacilando.org> Sent: Thursday, December 06, 2007 8:55 AM Subject: Re: [development] drupal.org caching too aggressive?
The APC issue doesn't apply here. It is a problem with the apc_cache module which is still alpha code and not ready for use (would be nice if someone would get interested in it and start working on it!) -Robert [...]
On Dec 6, 2007 5:32 AM, David Cohen <drupal@dave-cohen.com> wrote:
In case it helps, when this happens I never see the confirmation message which comes from drupal_set_message(). Which indicates something is happening to $_SESSION data. Usually, when a drupal_set_message() is not seen right away, it will appear eventually (next time a page is viewed with that session still active).
-Dave
This is the behavior I've seen a couple times.
On Wed, 5 Dec 2007 19:40:50 -0800, "Narayan Newton" <nnewton@osuosl.org> said:
I'll be looking at this, but I really doubt its replication delay at this point...
If it is not replication, we have a serious bug someplace.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Chris Johnson schrieb:
On Dec 6, 2007 5:32 AM, David Cohen <drupal@dave-cohen.com> wrote:
In case it helps, when this happens I never see the confirmation message which comes from drupal_set_message(). Which indicates something is happening to $_SESSION data. Usually, when a drupal_set_message() is not seen right away, it will appear eventually (next time a page is viewed with that session still active).
-Dave
This is the behavior I've seen a couple times.
On Wed, 5 Dec 2007 19:40:50 -0800, "Narayan Newton" <nnewton@osuosl.org> said:
I'll be looking at this, but I really doubt its replication delay at this point...
If it is not replication, we have a serious bug someplace.
This issue has been addressed. If it still happens, open an issue in the infra queue. Cheers, Gerhard -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFHXrinfg6TFvELooQRAp8JAKCNJlz7+3HDUiLVPRBmfUkmVj3jZQCfeVI5 4a+IvO+PdZHpU/nXoM4uY84= =2uKR -----END PGP SIGNATURE-----
On Dec 5, 2007 8:26 PM, Dave Cohen <drupal@dave-cohen.com> wrote:
Recently I've posted comments on Drupal.org, and after pressing the submit button I've seen the original post, without a message and without my comment. If I refresh, the comment is there. But at least once I've gone back and resubmitted, because as a user it looks like my post was simply ignored.
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on? Anyone else notice this?
I noticed this and noticed quite a few duplicate posts due to this behavior. Some patch submitters had their patches submitted three times, because they thought it did not get through. I found that it does not always show the post on refresh either. Possibly a side-effect of how our database replication works. Gabor
I've seen that a lot lately in comments to issues. I had a 6 minute delay before my comment appeared. -- begin storytime -- In GHOP issue 58, I replied to corsix about revisions needed for a patch, then posted my comment. As it turns out, he had posted a comment with a new patch before I even started adding my comment, and it didn't show until after I refreshed the issue a couple times because I was looking for my issue to appear. His patch even resolved the issues I was commenting about in the first place. -- end storytime -- If you ask me, whatever's going on is starting to become whatever the next stage is beyond slightly annoying. Dave Cohen wrote:
Recently I've posted comments on Drupal.org, and after pressing the submit button I've seen the original post, without a message and without my comment. If I refresh, the comment is there. But at least once I've gone back and resubmitted, because as a user it looks like my post was simply ignored.
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on? Anyone else notice this?
-Dave
On Wednesday 05 December 2007 11:26:25 Dave Cohen wrote:
Recently I've posted comments on Drupal.org, and after pressing the submit button I've seen the original post, without a message and without my comment. If I refresh, the comment is there. But at least once I've gone back and resubmitted, because as a user it looks like my post was simply ignored.
I suspect that I'm seeing a cached page after posting the comment (although I am logged in). Is this what's going on? Anyone else notice this?
-Dave
What I've taken to doing, since I noticed this, is to open a second browser to verify before reposting. For example, if I'm using Konqueror (usually), I'll open up Firefox. I'm not convinced that would always work, if it's a database replication issue as Derek says, but it has so far. Also, another thing I've noticed is that if I'm replying to an issue and I modify the issue's settings, they will always show, even if my comment does not (which really confused me the first time), indicating that it went through. -- Jason Flatt http://www.oadaeh.net/ Father of Six: http://www.flattfamily.com/ (Joseph, 14; Cramer, 12; Travis, 10; Angela; Harry, 7; and William, 12:04 am, 12-29-2005) Linux User: http://www.xubuntu.org/ Drupal Fanatic: http://drupal.org/
participants (12)
-
Chris Johnson -
Dave Cohen -
David Cohen
-
David Norman -
Derek Wright -
FGM -
Gerhard Killesreiter -
Gábor Hojtsy -
Jason Flatt -
Narayan Newton -
Robert Douglass -
Tomas J. Fulopp - Vacilando.org