[Osmf-talk] API Failover
penorman at mac.com
Mon Oct 27 21:59:24 UTC 2014
On 10/27/2014 9:21 AM, Joseph Reeves wrote:
> As someone involved in this original twittering, I didn't read the
> message like that. Sure $150k is a lot of money to keep the DB
> available during a planned reboot, but it would appear to be a very
> small amount of money to spend in order to keep the OSM database
> online if something catastrophic was to happen to the server it
> currently lives on. This weekend's interruption was brief; there has
> been larger interruptions in the past and there's always the
> possibility that Something Bad is going to unexpectedly happen.
There are redundancies in place to cover the case of a server failure of
one of the database servers. I wasn't involved in the planning of the
recent database server switchover, but I know it is substantially easier
to switch master and slave database servers if you can switch to read-only.
> I'm sure that database replication could also lead to some powerful
> use cases that aren't available to us yet: Performance improvements in
> places far from London, perhaps.
Having done a GSoC project on software that could be used for
distributed read-only API mirrors, I'd say the issues holding them back
are not hardware funding related, but time related. Having also run a
read-only API mirror, I know first-hand that it's actually a lot less
effective than you might think, particularly from North America.
No special access to the servers is required to run a read-only API
mirror - all the data is publicly available.
This is something that would be reasonable for an organization like HOT
to fund, or if someone wanted OSMF blessing for a grant proposal, I
would expect the board would indicate approval if it would help get funding.
The technical details aren't really a matter for osmf-talk@, but if
someone wants to discuss it in more detail, I'm probably the person most
familiar with it, or the dev@ mailing list is suitable.
More information about the osmf-talk