Hi all,

It looks like my long-awaited fix for some fundamental caching bugs is nearly ready to go to production. The code is done and reviewed by Alex so I'm only waiting on one more person and it will be ready to go. This change will fix a bunch of issues where following/ notifications are either missing or incorrect. Alex is also working on some code (http://twitter.com/al3x/status/1313993905) that will clean up even more issues. Hopefully that will fix these issues in the process.

Thanks;
  — Matt Sanford

On Mar 11, 2009, at 10:18 PM, dacort wrote:


I'll definitely post if I come across more. It's usually just more of
a Charlie Brown *augh* than a show-stopper by any means. And then like
you said, it eventually sorts itself out.

I did consider switching to XML as that always seems to be the cache
copy that's up-to-date, but then realized I would have to update a
variety of other things to deal with the proper data types.

Thanks for the detail!

dpc

On Mar 11, 11:05 am, Alex Payne <[email protected]> wrote:
Snapshots in time of the XML output and JSON output are helpful. But I certainly believe that there could be such discrepancies. They tend to
arise when our database replication lag increases. Some client
requests UncachedUserA in XML, and those inaccurate counts get "frozen in time" until a cache invalidation event occurs. Then the replication
lag sorts itself out, and someone requests UncachedUserB and sees the
correct values.

It's a very frustrating property of our system's architecture at the
moment, but one we're working on solutions for.



On Wed, Mar 11, 2009 at 10:09, Damon C <[email protected]> wrote:

I've seen this from time-to-time before, but could never really pin it
down to specific users or timeframes.

In any case, I frequently see inconsistencies between the data in XML
and JSON representations. The example I came across last night was
with the ActiveRainMaker account. The caches have since caught up (un)
fortunately, and I don't have another example.

In this case, though, the XML had a friends_count of ~1800 while the
JSON had a friends_count of ~800. I'm not really sure what else I can
provide and I probably didn't post before b/c it just sounds like
complaining. ;)

Thanks!

dpc

--
Damon P. Cortesi
Security Guy, Twitter Apps
www. tweetstats | tweepsearch | tweetsum .com

--
Alex Payne - API Lead, Twitter, Inc.http://twitter.com/al3x

Reply via email to