Dev: Add a function not to grant a specific ticket at a specific site

[inouekazu]

This request is a proposal patch about exclusion of the ticket, adds a function which does not grant that ticket to the site specified by exclude_site in booth.conf.

The example for which this function is needed.

• There are three sites. I want to operate two services (two tickets).
• However, I want to operate only one service at one site.

example of configuration

site="192.168.203.101"
site="192.168.203.102"
site="192.168.203.103"

ticket="ticketA"
        exclude_site = 192.168.203.102
ticket="ticketB"
        exclude_site = 192.168.203.101

• Then, I grant ticketA to 192.168.203.101 and ticketB to 192.168.203.102.
• By doing so, only 192.168.203.103 becomes a standby site of ticketA and ticketB.

[dmuhamedagic]

On Fri, May 30, 2014 at 03:18:05AM -0700, Kazunori INOUE wrote:

This request is a proposal patch about exclusion of the ticket, adds a function which does not grant that ticket to the site specified by exclude_site in booth.conf.

The example for which this function is needed.

• There are three sites. I want to operate two services (two tickets).
• However, I want to operate only one service at one site.

Looks like reasonable configuration. I'll comment more on the
patch a bit later.

[phmarek]

A simpler way would be to have 2 instances of booth; one with the arbitrator at A, and one with arbitrator B.

Regarding the patch: if you really want to do that,

• use a bitfield (like the existing site_bits/bitmask) for "allowed" sites within each ticket (initialized to -1, exclude_site removes bits, after reading the config mask all ticket masks again with the site_bits mask)
• use existing functions, like find_site_by_name, instead of having another strcmp loop
• at least some of the tests type == SITE will have to be changed to use that per-ticket bitfield now. Perhaps wrap that in an inline function.

That means there's no need for realloc or anything like that - just another value to be initialized, updated, and checked.

[dmuhamedagic]

On Mon, Jun 02, 2014 at 09:55:26AM -0700, phmarek wrote:

A simpler way would be to have 2 instances of booth; one with the arbitrator at A, and one with arbitrator B.

Yes, that's obviously another solution for this configuration.
Perhaps even preferable to adding another ticket option.
Kazunori-san, would that work for you too?

Regarding the patch: if you really want to do that,

• use a bitfield (like the existing site_bits/bitmask) for "allowed" sites within each ticket (initialized to -1, exclude_site removes bits, after reading the config mask all ticket masks again with the site_bits mask)

Yes, that's what I also wanted to write today.

• use existing functions, like find_site_by_name, instead of having another strcmp loop
• at least some of the tests type == SITE will have to be changed to use that per-ticket bitfield now. Perhaps wrap that in an inline function.

Probably right. I'm also not sure whether the change in
new_election() would suffice and was about to verify the rest of
the code.

[inouekazu]

Kazunori-san, would that work for you too?

I confirmed that booth had the function we wanted.
So I withdraw this request. Thanks!