[Tilesathome] Corrupted data in TRAPI database

[Jeremy Adams]

>
> It depends.  If the changeset is uploaded in small chunks to the API, each
> is written as a separate database transaction and the data will be spread
> across multiple minute diffs.  If the data is uploaded in a single chunk it
> is written within a single database transaction and can possibly take longer
> than 30 minutes to process.  The data only becomes visible to Osmosis after
> the transaction is committed by which time Osmosis has moved past that time
> interval and misses the data entirely.
>
>
>>
>> The fix seems to be using the new replication diffs, but TRAPI doesn't
>> support those yet.
>>
>
> Yes, they don't use timestamps at all so don't suffer from problems with
> long running transactions.  If you see any missing data in those files,
> please let me know.
>
>

Say I open a changeset and it takes an hour to upload.  Is the data in the
changeset being written bit by bit as I'm uploading for that whole hour, or
is it stored until I'm done uploading and then written to the database as
one large transaction?  If it's the latter, then I guess it doesn't matter
how long the changeset is open, only how long it takes the server to write
the changes to the database.

Either way, the solution seems to be using the new replication diffs.  I'm
no perl wizard, but I might try to muddle my way through and make some
changes.

-Jeremy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tilesathome/attachments/20091130/3c2273d8/attachment.html>