Huge performance decrease on Next() in latest versions about goracle HOT 4 CLOSED

Strange. Only 888df3d changed Next() between aa4eaea and 7a9e6c2.

If you can create a benchmark/test function for it, that would be the best.
Till that, first I'd comment out all the debug messages in Next, then look
around in fetchOneInto, maybe the reflect changes introduced such
performance degradation.

If you're adventureous enough, you can try out gopkg.in/rana/ora.v2

https://onename.com/tgulacsi

2015-08-26 15:47 GMT+02:00 Ricardo Smania [email protected]:

I was using aa4eaea
aa4eaea
, and when I updated to 7a9e6c2
7a9e6c2
I noticed a huge performance hit.

To test I used the database/sql compatible driver, querying and fetching 1
million records. I'm using the same machine and database, just changing the
repository between tests.

Old version: 56k rows/second
New version: 30k rows/second

I replaced the Next() func
https://github.com/go-goracle/goracle/blob/v1/driver.go#L225-L245 with
the old one and the performance increased dramatically, but it is still a
little slower than the old one, so I suspect something else is also to
blame:
New version with old Next(): 48k rows/second

—
Reply to this email directly or view it on GitHub
#8.

from goracle.

@tgulacsi I'm pretty sure 888df3d caused it.

When I have some spare time I'll create a benchmark function and try to investigate it.

I have tried Rana, but the database/sql implementation only reads 11.5k rows/sec. Goracle reads 56k (before 888df3d), and go-oci8 52k.

But Rana's specific Oracle implementation looks very promising. Right now I'm thinking about using Goracle for generic database/sql and Rana for PL/SQL, sys_refcursors and such. Like you said in your readme, it does have a better syntax. How about joining forces and making a single driver with the best of both? =)

from goracle.

I've contributed some to rana/ora:
https://github.com/rana/ora/commits/master :)

If you can provide us a good benchmark function, we can try to optimize
reading of rows.

https://onename.com/tgulacsi

2015-08-26 17:50 GMT+02:00 Ricardo Smania [email protected]:

@tgulacsi https://github.com/tgulacsi I'm pretty sure 888df3d
888df3d
caused it.

When I have some spare time I'll create a benchmark function and try to
investigate it.

I have tried Rana, but the database/sql implementation only reads 11.5k
rows/sec. Goracle reads 56k (before 888df3d
888df3d),
and go-oci8 52k.

But Rana's specific Oracle implementation looks very promising. Right now
I'm thinking about using Goracle for generic database/sql and Rana for
PL/SQL, sys_refcursors and such. Like you said in your readme, it does have
a better syntax. How about joining forces and making a single driver with
the best of both? =)

—
Reply to this email directly or view it on GitHub
#8 (comment).

from goracle.

@ricsmania I've just commited some performance improvement into gopkg.in/rana/ora.v3.
If you're still interested, take a look! It shaves off some time for selecting strings.

from goracle.