Build 2174 is out, and is running in our prod system

[Oren Eini (Ayende Rahien)]

Well, I hope that this should alleviate the memory issues that we have been seeing:

http://hibernatingrhinos.com/builds/ravendb-unstable-v2.0/2174-unstable for the details (and thanks for all the people who contributed).

This build also contains additional optimizations for map/reduce.

Like the previous build, on very large dbs, it may have a significant startup time for the first time. 

We would appreciate your feedback on the performance and stability of this release.

[Iulian Margarintescu]

It has only been 30-35minutes since i've deployed 2174 instead of 2173 but the difference is huge. Memory consumption is normal ( actually below normal ) cpu is idle, and indexes are up to date and only get stale for short periods of time. Also overall noticeable performance is a lot better than 2170 and 2173. The db has ~250k documents some larger ( 100-150kb ) some smaller ( <20kb). Hopefully it will continue running smoothly. 

Thanks for all the great work,

Iulian

[Oren Eini (Ayende Rahien)]

We actually noticed that there are still issues there, but I am happy that we are moving in the right direction.

[Paul Hinett]

Eager to test this out, but worried about the startup time.

Does the startup time only affect map/reduce indexes, i can probably
remove some of my map/reduce indexes to improve startup time this time
around if so.

Paul

On 19 December 2012 11:09:24, Oren Eini (Ayende Rahien) wrote:
> We actually noticed that there are still issues there, but I am happy
> that we are moving in the right direction.
>
> On Wed, Dec 19, 2012 at 1:07 PM, Iulian Margarintescu

> <eti...@gmail.com <mailto:eti...@gmail.com>> wrote:
>
> It has only been 30-35minutes since i've deployed 2174 <tel:2174>
> instead of 2173 <tel:2173> but the difference is huge. Memory

> consumption is normal ( actually below normal ) cpu is idle, and
> indexes are up to date and only get stale for short periods of
> time. Also overall noticeable performance is a lot better than

> 2170 <tel:2170> and 2173 <tel:2173>. The db has ~250k documents

> some larger ( 100-150kb ) some smaller ( <20kb). Hopefully it will
> continue running smoothly.
>
> Thanks for all the great work,
> Iulian
>
>
>
>

> On Wednesday, December 19, 2012 <tel:2012> 10:48:13 AM UTC+2, Oren

> Eini wrote:
>
> Well, I hope that this should alleviate the memory issues that
> we have been seeing:
>

> http://hibernatingrhinos.com/__builds/ravendb-unstable-v2.0/__2174-unstable

[Paul Hinett]

Iulian....what was your average memory consumption before the upgrade
to 2174 and what is it now?

How many map/reduce indexes do you use?

What was the initial startup time after updating to 2174?

Thanks!

[Iulian Margarintescu]

With 2173 mem usage was ~8-12Gb (machine has 16Gb). With 2174 mem usage is 3-5Gb. 

I have 6 map/reduce indexes and 4 normal indexes and the initial startup time was about a few minutes until the service was accessible and indexes were non stale ( for about 250K documents ).  

The memory usage seem to have risen now to about 6-7Gb. 

Iulian

[Oren Eini (Ayende Rahien)]

Yes, it map/reduce bound

[Paul Hinett]

I removed a couple of my map/reduce indexes on my dev machine and
upgraded, and amazingly it started up in about 30-60 seconds...a
dramatic difference from the 19 hours on the last schema update.

I feel more comfortable now and will update our production DB some
point today.

Disabling pre-fetching on 2173 seemed to work at first, but then
memory/cpu patterns seem to return to the same as before, so fingers
crossed the optimizations have done the job this time, i can see a lot
of work went into this build...thanks!

> http://hibernatingrhinos.com/____builds/ravendb-unstable-v2.0/____2174-unstable
> <http://hibernatingrhinos.com/__builds/ravendb-unstable-v2.0/__2174-unstable>

[Oren Eini (Ayende Rahien)]

Paul

I'm pretty sure that still isn't it

Or our system we see something similar

We continue to incestigate