[Tilesathome] Tile retries

[spaetz]

Xaa has been looking into osmarender to find one of the crash reasons. As 80n already told, the path length lookup recurses too often for some long ways and Xaa believes he might come up with a patch that aborts that recursion as soon as our way is long enough for the text being rendered.

On a related issue, a lot of clients don't seem to return results in time (for whatever reasons), currently our queue looks like this:

+--------+----------+
| status | count(*) |
+--------+----------+
|pending |     5905 | 
|retry   |     2339 | 

Jobs time out after some hours and are requeued with the timeout date as "job insertion date", so they basically start again at the very beginning of the queue.
But still, this number is way too high.

Some of the tiles that have repeatedly failed to render:
+------+------+---------+
| x    | y    | retries |
+------+------+---------+
| 2392 |  901 |       5 | 
| 1044 | 1550 |       5 | 
| 1173 | 1562 |       5 | 
| 1174 | 1561 |       5 | 
| 1175 | 1558 |       5 | 
| 1175 | 1560 |       5 | 
| 1176 | 1559 |       5 | 
| 1177 | 1559 |       5 | 
| 1179 | 1555 |       5 | 
| 1170 | 1560 |       5 | 
| 2184 | 1137 |       6 | 
+------+------+---------+

I heard, London is among that area, so dense areas seem to be a problem. We do need to look into osmarender related crashes as well as into inkscape related problems. As well as better error reporting from the client side. currently a crashed tile will use up an "active job" slot until the job times out, which is obviously not good. therefore many crashes lead to less tiles rendered.

spaetz