From: Junio C Hamano <gitster@pobox.com>
To: Stefan Beller <sbeller@google.com>
Cc: git@vger.kernel.org, jacob.keller@gmail.com, peff@peff.net,
jrnieder@gmail.com, johannes.schindelin@gmail.com,
Jens.Lehmann@web.de, vlovich@gmail.com
Subject: Re: [PATCHv3 06/13] run-command: add an asynchronous parallel child processor
Date: Mon, 21 Sep 2015 18:08:33 -0700 [thread overview]
Message-ID: <xmqqfv276z1q.fsf@gitster.mtv.corp.google.com> (raw)
In-Reply-To: <1442875159-13027-7-git-send-email-sbeller@google.com> (Stefan Beller's message of "Mon, 21 Sep 2015 15:39:12 -0700")
Stefan Beller <sbeller@google.com> writes:
> +void default_start_failure(void *data,
> + struct child_process *cp,
> + struct strbuf *err)
> +{
> + int i;
> + struct strbuf sb = STRBUF_INIT;
> +
> + for (i = 0; cp->argv[i]; i++)
> + strbuf_addf(&sb, "%s ", cp->argv[i]);
> + die_errno("Starting a child failed:\n%s", sb.buf);
Do we want that trailing SP after the last element of argv[]?
Same question applies to the one in "return-value".
> +static void run_processes_parallel_init(struct parallel_processes *pp,
> + int n, void *data,
> + get_next_task_fn get_next_task,
> + start_failure_fn start_failure,
> + return_value_fn return_value)
> +{
> + int i;
> +
> + if (n < 1)
> + n = online_cpus();
> +
> + pp->max_processes = n;
> + pp->data = data;
> + if (!get_next_task)
> + die("BUG: you need to specify a get_next_task function");
> + pp->get_next_task = get_next_task;
> +
> + pp->start_failure = start_failure ? start_failure : default_start_failure;
> + pp->return_value = return_value ? return_value : default_return_value;
I would actually have expected that leaving these to NULL will just
skip pp->fn calls, instead of a "default implementation", but a pair
of very simple default implementation would not hrtut.
> +static void run_processes_parallel_cleanup(struct parallel_processes *pp)
> +{
> + int i;
Have a blank between the decl block and the first stmt here (and
elsewhere, too---which you got correct in the function above)?
> + for (i = 0; i < pp->max_processes; i++)
> + strbuf_release(&pp->children[i].err);
> +static void run_processes_parallel_start_one(struct parallel_processes *pp)
> +{
> + int i;
> +
> + for (i = 0; i < pp->max_processes; i++)
> + if (!pp->children[i].in_use)
> + break;
> + if (i == pp->max_processes)
> + die("BUG: bookkeeping is hard");
Mental note: the caller is responsible for not calling this when all
slots are taken.
> + if (!pp->get_next_task(pp->data,
> + &pp->children[i].process,
> + &pp->children[i].err)) {
> + pp->all_tasks_started = 1;
> + return;
> + }
Mental note: but it is OK to call this if get_next_task() previously
said "no more task".
The above two shows a slight discrepancy (nothing earth-breaking).
I have this suspicion that the all-tasks-started bit may turn out to
be a big mistake that we may later regret. Don't we want to allow
pp->more_task() to say "no more task to run at this moment" implying
"but please do ask me later, because I may have found more to do by
the time you ask me again"?
That is one of the reasons why I do not think the "very top level is
a bulleted list" organization is a good idea in general. A good
scheduling decision can seldom be made in isolation without taking
global picture into account.
> +static void run_processes_parallel_collect_finished(struct parallel_processes *pp)
> +{
> + int i = 0;
> + pid_t pid;
> + int wait_status, code;
> + int n = pp->max_processes;
> +
> + while (pp->nr_processes > 0) {
> + pid = waitpid(-1, &wait_status, WNOHANG);
> + if (pid == 0)
> + return;
> +
> + if (pid < 0)
> + die_errno("wait");
> +
> + for (i = 0; i < pp->max_processes; i++)
> + if (pp->children[i].in_use &&
> + pid == pp->children[i].process.pid)
> + break;
> + if (i == pp->max_processes)
> + /*
> + * waitpid returned another process id
> + * which we are not waiting for.
> + */
> + return;
If we culled a child process that this machinery is not in charge
of, waitpid() in other places that wants to see that child will not
see it. Perhaps such a situation might even warrant an error() or
BUG()? Do we want a "NEEDSWORK: Is this a bug?" comment here at
least?
> + if (strbuf_read_once(&pp->children[i].err,
> + pp->children[i].process.err, 0) < 0 &&
> + errno != EAGAIN)
> + die_errno("strbuf_read_once");
Don't we want to read thru to the end here? The reason read_once()
did not read thru to the end may not have anything to do with
NONBLOCK (e.g. xread_nonblock() caps len, and it does not loop).
next prev parent reply other threads:[~2015-09-22 1:08 UTC|newest]
Thread overview: 48+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-09-21 22:39 [PATCHv3 00/13] fetch submodules in parallel and a preview on parallel "submodule update" Stefan Beller
2015-09-21 22:39 ` [PATCHv3 01/13] Sending "Fetching submodule <foo>" output to stderr Stefan Beller
2015-09-21 23:47 ` Junio C Hamano
2015-09-21 22:39 ` [PATCHv3 02/13] xread: poll on non blocking fds Stefan Beller
2015-09-21 23:55 ` Junio C Hamano
2015-09-22 4:55 ` Torsten Bögershausen
2015-09-22 6:23 ` Jacob Keller
2015-09-22 18:40 ` Torsten Bögershausen
2015-09-22 19:45 ` Junio C Hamano
2015-09-22 19:49 ` Jeff King
2015-09-22 20:00 ` Junio C Hamano
2015-09-23 0:14 ` Stefan Beller
2015-09-23 0:43 ` Junio C Hamano
2015-09-23 1:51 ` Jeff King
2015-09-21 23:56 ` Eric Sunshine
2015-09-22 15:58 ` Junio C Hamano
2015-09-22 17:38 ` Stefan Beller
2015-09-22 18:21 ` Junio C Hamano
2015-09-22 18:41 ` Stefan Beller
2015-09-21 22:39 ` [PATCHv3 03/13] xread_nonblock: add functionality to read from fds nonblockingly Stefan Beller
2015-09-22 0:02 ` Junio C Hamano
2015-09-22 0:10 ` Junio C Hamano
2015-09-22 6:26 ` Jacob Keller
2015-09-22 6:27 ` Jacob Keller
2015-09-22 15:59 ` Junio C Hamano
2015-09-21 22:39 ` [PATCHv3 04/13] strbuf: add strbuf_read_once to read without blocking Stefan Beller
2015-09-22 0:17 ` Junio C Hamano
2015-09-22 6:29 ` Jacob Keller
2015-09-21 22:39 ` [PATCHv3 05/13] run-command: factor out return value computation Stefan Beller
2015-09-22 0:38 ` Junio C Hamano
2015-09-21 22:39 ` [PATCHv3 06/13] run-command: add an asynchronous parallel child processor Stefan Beller
2015-09-22 1:08 ` Junio C Hamano [this message]
2015-09-22 18:28 ` Stefan Beller
2015-09-22 19:53 ` Junio C Hamano
2015-09-22 21:31 ` Stefan Beller
2015-09-22 21:41 ` Junio C Hamano
2015-09-22 21:54 ` Stefan Beller
2015-09-22 22:23 ` Junio C Hamano
2015-09-21 22:39 ` [PATCHv3 07/13] fetch_populated_submodules: use new parallel job processing Stefan Beller
2015-09-22 16:28 ` Junio C Hamano
2015-09-21 22:39 ` [PATCHv3 08/13] submodules: allow parallel fetching, add tests and documentation Stefan Beller
2015-09-21 22:39 ` [PATCHv3 09/13] submodule config: keep update strategy around Stefan Beller
2015-09-22 0:56 ` Eric Sunshine
2015-09-22 15:50 ` Stefan Beller
2015-09-21 22:39 ` [PATCHv3 10/13] git submodule update: cmd_update_recursive Stefan Beller
2015-09-21 22:39 ` [PATCHv3 11/13] git submodule update: cmd_update_clone Stefan Beller
2015-09-21 22:39 ` [PATCHv3 12/13] git submodule update: cmd_update_fetch Stefan Beller
2015-09-21 22:39 ` [PATCHv3 13/13] Rewrite submodule update in C Stefan Beller
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=xmqqfv276z1q.fsf@gitster.mtv.corp.google.com \
--to=gitster@pobox.com \
--cc=Jens.Lehmann@web.de \
--cc=git@vger.kernel.org \
--cc=jacob.keller@gmail.com \
--cc=johannes.schindelin@gmail.com \
--cc=jrnieder@gmail.com \
--cc=peff@peff.net \
--cc=sbeller@google.com \
--cc=vlovich@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).