On 09/24/2011 05:46 PM, Kevin Grittner wrote:
>> Mikko Tiihonen wrote:
>
>> In general could add a mechanism to add/remove specific oids from
>> the binary transfer mode as a configuration option to the jdbc
>> driver.
>>
>> The same mechanism could be used for by application code that uses
>> custom types inside PostgreSQL to take advantage of binary
>> transfers.
>>
>> There will always be trade-offs between txt and bin encoding and
>> sometimes the differences are large enough to change from driver
>> defaults.
>
> Yeah, all I'm saying is that when there are such large trade-offs
> between CPU usage and network traffic requirements, it might be a
> good idea to allow a user to choose with some connection option. For
> perspective, just within our organization we have cases where we're
> communicating between busy machines on the same 1Gbps switch, and
> cases where we're communicating between nearly-idle machines over a
> sometimes-saturated 3Mbps WAN connection. This change would be a big
> win for one and could potentially be a big problem in the other.
Here is a patch that allows using binaryTransferEnable and
binaryTransferDisable options to allow driver used to override the
defaults.
It accepts a comma separated list of OID names (much match names in
org.postgres.core.Oid) or any numeric oid value.
For example one can use
binaryTransferDisable=INT4_ARRAY,1016
binaryTransferEnable=TEXT
It will also be useful if for some reason the binary encoding of a
variable changes in the server to allow using of old jdbc driver by
disabling the binary transfer for one type as a workaround while
waiting for a new jdbc driver release.
-Mikko
