Re: [HACKERS] [PATCHES] fork/exec patch

>> I think custom code will be needed. It could be in the form
>of a wrapper
>> or in the form of integrated code. But if you use one of the general
>> wrappers, it's going to be hard to teach it to send signals to the
>> postmaster in the way we want (regardless of which method is
>chosen for
>> signals)
>>
>
>Right. I expect we'll need 3 extra command line switches: one
>to install
>as a service, one to remove the service, and one to run as a
>service. Or
>possibly a small wrapper for the install/remove part, plus one extra
>command switch.

Normally, that would be handled by the installation program. But there'd
definitly be a need for a manual way to do it as well, yes. (There are
tools in the resource kit, but people shouldn't need that)

We'd probably need a tool like "pgsignal" or something that would send
signals to the backends. Since you can't use the kill commandline tool
(that one only supports what would be called kill -9 on unix)

>> But it's probably best to just get it up and running on a
>command prompt
>> first :-)
>>
>
>Indeed.
>
>Another question: do we plan to use the Windows event log
>instead of or as well as syslog?

Well, definitly at least as well as.
AFAIK, there is no builtin syslog support in Win32 (why should there
be), so if we want to provide both, we'd need to either implement the
syslog client ourselves, or find some good place to borrow it from.
Question is, would anyone use it. The evenlog is what windows people are
used to using, and any large shop will have things set up for handling
them.
Based on this, I'd vote instead of. But if we can do it at little extra
cost (e.g. there is a good syslog client we can borrow), then why not do
both as long as it's cheap.

//Magnus