Probability distribution of eventual consistency

78 views
Skip to first unread message

Jeff Schnitzer

unread,
Jul 28, 2015, 1:30:11 PM7/28/15
to Google App Engine
Is there any kind of published document that describes the probability curve of eventual consistency in the HRD? Say, probability of query showing results on Y axis, time on X axis. I would expect a steep S.

Googlers: I suspect you have something like this this plotted internally. It would really help us developers determine whether certain kinds of grey-zone queries can live with eventuality or whether we need to adapt our schemas. It doesn't have to be exact; I'd be happy with ballpark figures.

Thanks,
Jeff

Joshua Smith

unread,
Jul 28, 2015, 2:21:47 PM7/28/15
to google-a...@googlegroups.com
I’ve never seen it take longer than a second, except when it was taking several hours during an outage that one time.

So I guess the graph looks like this:

/\_________|___
01        wtf?




--
You received this message because you are subscribed to the Google Groups "Google App Engine" group.
To unsubscribe from this group and stop receiving emails from it, send an email to google-appengi...@googlegroups.com.
To post to this group, send email to google-a...@googlegroups.com.
Visit this group at http://groups.google.com/group/google-appengine.
To view this discussion on the web visit https://groups.google.com/d/msgid/google-appengine/CADK-0ui6JfSU9t4XYiR6-GOEoCCvA6Ce9JpogmRM43NDz%2B5eNg%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Patrice (Cloud Platform Support)

unread,
Jul 30, 2015, 10:42:51 AM7/30/15
to Google App Engine, mrjoshu...@gmail.com, mrjoshu...@gmail.com
Hi Jeff,

I did some digging around with the datastore team, and turns out this is not in any form that is available publicly. 

The next best thing I can provide was the answer from the team when I asked for "ballpark", and it was "it should never take more than a few seconds", so I'd say Joshua's graph, while very minimal, is close to the truth. Under a few seconds, consistency should be applied on your datastore.

I hope that this is enough data, even if it's not a graph.

Cheers

Nick

unread,
Jul 30, 2015, 6:04:20 PM7/30/15
to Google App Engine
What about consistency in undocumented services, such as the search API, or querying pull queues by tag? What would the team say the timeframes for those are?

Patrice (Cloud Platform Support)

unread,
Jul 31, 2015, 12:40:28 PM7/31/15
to Google App Engine, naok...@gmail.com
Hi Nick,

This is a completely different question, for a completely different team. I'd surmise that if we don't have a graph for one of the most common and documented service, that we won't have more information on less documented one.

In any case, the fact that these are not documented is that, even for ballpark figures, this is a question that is extremely hard to answer, as it is dependent on such a number of heuristics that, even though it's fast, it's hard to give a more precise number.

Cheers
Reply all
Reply to author
Forward
0 new messages