Re: Fluidinfo performance
Terry Jones
Sorry for the slow reply.
> 1. Is "has" operator faster then "matches" ? What is the fastest way to
> query fluidinfo ?
I'll leave Esteve or Manuel to answer that.
> 2. How can we implement pagination on these queries, we don't want to
> load 1mil results ? We need a LIMIT sort of operator
Yes, agreed. This is under active discussion right now.
> 3. How many simultaneous clients can query the system ? Ideally, we want
> O(1) query complexity on 1000 parallel queries .
Well, O(1) includes "a week" or any other constant, so I think we've got
you covered already :-)
Right now we have requests going through haproxy and there are 8 processes
handling queries, I think up to 2 simultaneously. After that things will be
queued momentarily. We can obviously increase that. Jamu is probably the
best person to comment on this.
> 4. I know that objects cannot be deleted, but what happens if we remove
> the tags and a lot of dummy objects will be kept in the system ?
Yes, but empty objects don't take up any space, apart from one row in a db
(it's only tags that cause storage). So I certainly wouldn't worry about
that at all, but thanks!
Terry
Jamu Kakar
On Wed, Feb 29, 2012 at 6:22 PM, Terry Jones <te...@jon.es> wrote:
>> 1. Is "has" operator faster then "matches" ? What is the fastest way to
>> query fluidinfo ?
>
> I'll leave Esteve or Manuel to answer that.
If you want everything with a particular tag then use 'has'. If you
want fuzzy text matching then use 'matches'.
>> 3. How many simultaneous clients can query the system ? Ideally, we want
>> O(1) query complexity on 1000 parallel queries .
>
> Well, O(1) includes "a week" or any other constant, so I think we've got
> you covered already :-)
>
> Right now we have requests going through haproxy and there are 8 processes
> handling queries, I think up to 2 simultaneously. After that things will be
> queued momentarily. We can obviously increase that. Jamu is probably the
> best person to comment on this.
We have 8 API services load balanced by HAProxy, as Terry says;
however, they only handle a single request at a time. HAProxy keeps
track of busy services and passes any queued requests to the first
available API service. We can increase this number quite easily, but
so far performance with 8 API services seems to be pretty good.
Thanks,
J.