Hi,
after returning from my canada holidays, I see that the tirex "dirty" rendering queue is inactive. Has it been deactivated by purpose or is this an error?
Tirex Master Status (updated=2011-08-21 11:06:12)
Master server: started=2011-08-10 21:36:48 pid=17815
Queue: Prio Size Maxsize Age 1 0 233 2 151144 151144 8792:53-14495:54 10 37 42 30:42-14204:24
all 151181 151181 30:42-14495:54
Buckets: (load=1.61) Name Priority Rend. MaxRd Maxld Active Can Queued Age missing 1- 1 3 6 20 yes yes 0 *dirty 2- 9 0 4 8 no no 151144 8792:53-14495:54 bulk 10- 19 0 3 6 yes no 37 30:42-14204:24 background 20- 0 3 4 yes no 0
^see line with the * Active=no
Kind regards, Kay
Am 21.08.2011 13:22, schrieb Kay Drangmeister:
Hi,
after returning from my canada holidays, I see that the tirex "dirty" rendering queue is inactive. Has it been deactivated by purpose or is this an error?
someome suggested to turn down rendering until the database has catched up. I did this by disabling the dirty queue, so that the missing tiles are still rendered.
Peter
The plan is that if we catched up we will reactivated the expire-script and begin than to render all tiles z=9-15 systematically, after this we will rendering systematically all tiles that are older than a year (expired) and repeat this if we are done. I'm not sure what we do with z=16-18, I would say we do this like in the past.
Greetings Tim
Am 21.08.2011 13:24, schrieb Peter Körner:
someome suggested to turn down rendering until the database has catched up. I did this by disabling the dirty queue, so that the missing tiles are still rendered.
Peter
Hello, we have now an up-to-date database, and expiring is working again. I start to rendering all tiles z=9-18 systematically, that are watched in the last 64 days and older than 3 days (because expire didn't work a long time). I put additional z=16-18 in the batch because it is else difficult to handle in different processes.
We rendering 150-200 tiles/min what is really nice.
There are 3 Mio. tiles in the batch so it will need around 11 days.
After this we need to re-render only tiles that are older than a year because expiring is again working, so there will be than much fewer tiles to handle.
Greetings Kolossos
Am 21.08.2011 14:11, schrieb Tim Alder:
The plan is that if we catched up we will reactivated the expire-script and begin than to render all tiles z=9-15 systematically, after this we will rendering systematically all tiles that are older than a year (expired) and repeat this if we are done. I'm not sure what we do with z=16-18, I would say we do this like in the past.
Greetings Tim
Am 21.08.2011 13:24, schrieb Peter Körner:
someome suggested to turn down rendering until the database has catched up. I did this by disabling the dirty queue, so that the missing tiles are still rendered.
Peter
Hi,
Am 27.08.2011, 23:11 Uhr, schrieb Tim Alder tim.alder@s2002.tu-chemnitz.de:
Hello, we have now an up-to-date database, and expiring is working again.
Looks great. http://toolserver.org/~mazder/tirex-status/?short=1&extended=0&refre... shows a new queue "systema", I guess that's exactly it.
One observation: currently putting re-rendered tiles into the queue is a bit quicker than the queue rendering tiles, so the queue grows slowly over time. Nothing to worry I guess, tho'. But: should the AGE of the tiles not gradually decrease, as the queue is supposed to render oldest tiles first?
And is "dirty" now separate from the "systema" queue (as is apparent)? If I try to manually dirty a metatile, I don't see it appearing anywhere on the page linked above.
Kind regards, Kay
Am 27.08.2011 23:31, schrieb Kay Drangmeister:
... But: should the AGE of the tiles not gradually decrease, as the queue is supposed to render oldest tiles first?
No, as far I know the queue is only order by numbers of requests.
And is "dirty" now separate from the "systema" queue (as is apparent)? If I try to manually dirty a metatile, I don't see it appearing anywhere on the page linked above.
Yes, that's really still a problem. I didn't find a trick how to set dirty-tiles to prio=3 or to deactivated the dirty-bucket and push the list under "bulk" (prio=10) into tirex.
Greetings Kolossos
Am 27.08.2011 23:31, schrieb Kay Drangmeister:
Hi,
Am 27.08.2011, 23:11 Uhr, schrieb Tim Aldertim.alder@s2002.tu-chemnitz.de:
Hello, we have now an up-to-date database, and expiring is working again.
Looks great. http://toolserver.org/~mazder/tirex-status/?short=1&extended=0&refre... shows a new queue "systema", I guess that's exactly it.
One observation: currently putting re-rendered tiles into the queue is a bit quicker than the queue rendering tiles, so the queue grows slowly over time. Nothing to worry I guess, tho'. But: should the AGE of the tiles not gradually decrease, as the queue is supposed to render oldest tiles first?
The queue is sorted by the number of requests received for this tile, not by age.
And is "dirty" now separate from the "systema" queue (as is apparent)? If I try to manually dirty a metatile, I don't see it appearing anywhere on the page linked above.
when tirex receives a dirty message from mod_tile (metatile oder then planet-timestamp), it enqueues it with prio 2, so it's put into the systema bucket. This can be changed in source [1] (i guess the second int in the array is the priority for cmdRender).
Also, if you enqueue a tile that is already in one of the queues, it's not added a second time but just moved up a little in the queue.
Peter
[1] http://trac.openstreetmap.org/browser/applications/utils/tirex/lib/Tirex/Source/ModTile.pm#L121
On 08/29/2011 02:22 AM, Peter Körner wrote:
Am 27.08.2011 23:31, schrieb Kay Drangmeister:
Hi,
Am 27.08.2011, 23:11 Uhr, schrieb Tim Aldertim.alder@s2002.tu-chemnitz.de:
Hello, we have now an up-to-date database, and expiring is working again.
Looks great. http://toolserver.org/~mazder/tirex-status/?short=1&extended=0&refre... shows a new queue "systema", I guess that's exactly it.
One observation: currently putting re-rendered tiles into the queue is a bit quicker than the queue rendering tiles, so the queue grows slowly over time. Nothing to worry I guess, tho'. But: should the AGE of the tiles not gradually decrease, as the queue is supposed to render oldest tiles first?
The queue is sorted by the number of requests received for this tile, not by age.
Will this not quickly mess up the systematic ordering of the systema queue? Is it possible to turn off this reordering?
Has this got something to do with that throughput has dropped from 160 - 200 down to (still better than before) 60 - 70 tiles / minute? But overall, it does seem the systematicity is working.
And is "dirty" now separate from the "systema" queue (as is apparent)? If I try to manually dirty a metatile, I don't see it appearing anywhere on the page linked above.
when tirex receives a dirty message from mod_tile (metatile oder then planet-timestamp), it enqueues it with prio 2, so it's put into the systema bucket. This can be changed in source [1] (i guess the second int in the array is the priority for cmdRender).
Perhaps the mappings from mod_tile render request types to tirex priority levels could be made configurable.
Alternatively, one could tune the mod_tile config to send requests as cmdDirty rather than cmdRender (the latter assumes the request can be rendered on the fly, which isn't the case if about 300k tiles are ahead of it in the queue) mod_tiles cmdDirty maps onto tirex prio level 10 by default.
Kai
P.S. In the run.log file, at the end of each import, there are messages of the form
26799 import done [2011-08-29 02:11:54] 26799 expiring tiles [2011-08-29 02:11:54] 26799 [error] tile_expiry error [2011-08-29 02:11:55] 26799 resetting state
Does the "resetting state" have anything to do with that the replag is not going down?
Also, if you enqueue a tile that is already in one of the queues, it's not added a second time but just moved up a little in the queue.
Peter
[1] http://trac.openstreetmap.org/browser/applications/utils/tirex/lib/Tirex/Source/ModTile.pm#L121
Maps-l mailing list Maps-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/maps-l
Am 29.08.2011 16:40, schrieb Kai Krueger:
Will this not quickly mess up the systematic ordering of the systema queue? Is it possible to turn off this reordering?
Has this got something to do with that throughput has dropped from 160 - 200 down to (still better than before) 60 - 70 tiles / minute?
Hello, my theory is that we started with rendering around Alaska. This area is relative empty and stretched by mercator-projection. Now we are somewhere in Canada/USA incl. NYC with high density of geo-objects. Following the z-curve we should come back to empty areas in the north several times, and render again very fast. If not, my theory is false.
In general: Tirex should be able to order the queue by z-curve, especially if you have more than one style (not sure how it can reach the end of it), it would makes it for me much easier to handle it. And Tirex should be able to not order the queue.
when tirex receives a dirty message from mod_tile (metatile oder then planet-timestamp), it enqueues it with prio 2, so it's put into the systema bucket. This can be changed in source [1] (i guess the second int in the array is the priority for cmdRender).
I change the parameter to 4 in ~/tirex/lib/vendor_perl/5.12/Tirex/Source/ModTile.pm It seems that it need a restart of tirex, what I don't want to do in the moment.
Alternatively, one could tune the mod_tile config to send requests as cmdDirty rather than cmdRender (the latter assumes the request can be rendered on the fly, which isn't the case if about 300k tiles are ahead of it in the queue) mod_tiles cmdDirty maps onto tirex prio level 10 by default.
No idea where to do this.
Kai
P.S. In the run.log file, at the end of each import, there are messages of the form
26799 import done [2011-08-29 02:11:54] 26799 expiring tiles [2011-08-29 02:11:54] 26799 [error] tile_expiry error [2011-08-29 02:11:55] 26799 resetting state
Does the "resetting state" have anything to do with that the replag is not going down?
Would be important to know. It seems to work at beginning.
Hi
Am 29.08.2011 21:36, schrieb Tim Alder:
P.S. In the run.log file, at the end of each import, there are messages of the form
26799 import done [2011-08-29 02:11:54] 26799 expiring tiles [2011-08-29 02:11:54] 26799 [error] tile_expiry error [2011-08-29 02:11:55] 26799 resetting state
Does the "resetting state" have anything to do with that the replag is not going down?
Would be important to know. It seems to work at beginning.
The load-next script had the expire.rb script enabled which is the wrong method of expiring. The correct one follows some lines after the expire.rb statement. I commented out the bad commands.
I also lowered the maxInterval setting from 24h to 6h, so we can check sooner if this fixed the problem.
Peter
Hello, we are rendering now somewhere in the middle of atlantic[1]. So I would like to know why we are not so fast as at the begining?
If I look in psql the most queries are <IDLE>. We are rendering now 60 Metatiles/minute, so we are not bad, but I would like to understand.
Greetings Tim
[1] http://openstreetmap.gryph.de/bigmap.cgi?zoom=14&xmin=7656&xmax=7664...
Am 21.08.2011 13:24, schrieb Peter Körner:
Am 21.08.2011 13:22, schrieb Kay Drangmeister:
tirex "dirty" rendering queue is inactive.
someome suggested to turn down rendering until the database has catched up. I did this by disabling the dirty queue,
Thanks for the explanation (is there a wiki page about the current status showing this?). When is the projected date/time the DB should have catched up? The age of the queue suggest it's been more than 11 days already... Is there a monitoring possibility?
Thanks, Kay
Hello Kay, in this mailing list you would find the answer. Peter wrote a little tool to show the replag: http://toolserver.org/~mazder/tirex-replag/?refresh=0&human=1 So we have now a replag of 4 days and I would say we need 3 days to be up-to-date.
Unfortunately is munin-graph again out of order: http://munin.toolserver.org/OSM/ptolemy/replication_delay2.html and jira-request TS-1125 after nearly a month still untouched. :-(
I believe it would be difficult for us to keep the wiki up-to-data, so let's use this list. But I know that's not the optimal medium if somebody is coming for a long vacation or so, so it's not the problem that you ask.
Greetings Kolossos
Am 22.08.2011 22:19, schrieb Kay Drangmeister:
Am 21.08.2011 13:24, schrieb Peter Körner:
Am 21.08.2011 13:22, schrieb Kay Drangmeister:
tirex "dirty" rendering queue is inactive.
someome suggested to turn down rendering until the database has catched up. I did this by disabling the dirty queue,
Thanks for the explanation (is there a wiki page about the current status showing this?). When is the projected date/time the DB should have catched up? The age of the queue suggest it's been more than 11 days already... Is there a monitoring possibility?
Thanks, Kay
Maps-l mailing list Maps-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/maps-l