[PATCH 3 of 3] worker: handle worker failures more aggressively
diptongo at gmail.com
Thu Feb 21 15:58:48 CST 2013
Replying Bryan O'Sullivan:
> On Thu, Feb 21, 2013 at 12:22 AM, Isaac Jurado <diptongo at gmail.com> wrote:
>> That's my point. Working with processes on Windows is going to be
>> different enough from the other platforms, right?
> Okay, I'm confused. Are you looking for me to do something, or just
> passing the time?
Last time I reply on this, I promise. Let me summarize what, from my
point of view, has happened.
- You submitted a patch that uses a monitor thread on the parent
process in order to detect a children failure early and finish the
rest of the children.
- I suggested to leave that task to the children processes by putting
all of them in the same process group and "broadcasting" a signal on
failure. Which could circumvent the need for threading the parent
- You replied that process groups are Unix only and you wanted to
maximize code portability.
- I was confused, since os.fork() and atomic writes to pipes are also
Unix only (POSIX in particular). So I tried to ask for a better
explanation, as politely as possible. Specially since mailing list
discussions tend to heat up very quickly and, between you and me,
you are the authority. I don't want any trouble.
- After a pair of replies, you told me, more or less, that Windows is
a whole different story. Statement that I found very contradictory
from the first argument rejecting the use of process groups.
Therefore, when, in theroy, I was trying to understand the resasonings behind some
design decisions, in practice I was unwillingly passing the time.
As I said, there's no point in taking this discussion further. Whatever
you write will be fine, even if apparently contradictory and if I have
any complaint, I'll send a patch.
Sorry for the confusion. Best regards.
"The noblest pleasure is the joy of understanding."
Leonardo da Vinci
More information about the Mercurial-devel