Re: [Patch] Do not crash upon remote-remote migration

Another keeper for the new official git tree.

Moshe



On 3/31/06, Ansgar Esztermann <[email protected]> wrote:
>
> When asking an already migrated process to move to another (remote)
> node, bring it back home first. The current implementation leads to a
> crash. A direct remote->remote migration would be better, but is not
> implemented yet.
>
> ---
>
> hpc/migctrl.c |   84
> +++++++++++++++++++++++++++++++--------------------------
> 1 files changed, 46 insertions(+), 38 deletions(-)
>
> 2a7d618eb6d2ae0942349f16e7a46d13351d7dc2
> diff --git a/hpc/migctrl.c b/hpc/migctrl.c
> index 18a798a..dee9adb 100644
> --- a/hpc/migctrl.c
> +++ b/hpc/migctrl.c
> @@ -79,6 +79,44 @@ int task_remote_wait_expel(task_t *p)
>         return task_remote_expel(p);
> }
>
> +
> +/**
> + * task_local_bring - Receive task back in the deputy stub
> + * @p:         deputy task to receive
> + * @reason:    reason to send (if any)
> + **/
> +static int task_local_bring(task_t *p, int reason)
> +{
> +       int error;
> +       om_link_t *link;
> +
> +       if (obtain_mm(p)) {
> +               printk(KERN_ERR "unable to obtain mm\n");
> +               goto failed;
> +       }
> +
> +       /* send remote request */
> +       comm_send_req(p->om.contact, DEP_COMING_HOME);
> +
> +       /* see if other part is with on this */
> +       if (mig_recv_hshake(p->om.contact))
> +               goto failed;
> +
> +       /* receive the process back */
> +       error = mig_do_receive(p);
> +       if (error)
> +               goto failed;
> +
> +       task_clear_dflags(p, DDEPUTY);
> +       link = task_set_comm(p, NULL);
> +       comm_close(link);
> +
> +       return 0;
> +failed:
> +       OMBUG("failed\n");
> +       return -1;
> +}
> +
> /**
>   * task_local_send - Send a local task to remote
>   * @p:         task to send
> @@ -90,6 +128,14 @@ static int task_local_send(task_t *p, st
>         om_link_t *mlink;
>         int error = 0;
>
> +       /* if this is a stub, bring it back first */
> +       if (p->om.contact)
> +       {
> +               error = task_local_bring(p, reason);
> +               if (error)
> +                       goto failed;
> +       }
> +
>         sockaddr_setup_port(whereto, REMOTE_DAEMON_PORT);
>         mlink = comm_setup_connect(whereto, 0);
>         if (!mlink) {
> @@ -123,44 +169,6 @@ failed:
> }
>
>
> -
> -/**
> - * task_local_bring - Receive task back in the deputy stub
> - * @p:         deputy task to receive
> - * @reason:    reason to send (if any)
> - **/
> -static int task_local_bring(task_t *p, int reason)
> -{
> -       int error;
> -       om_link_t *link;
> -
> -       if (obtain_mm(p)) {
> -               printk(KERN_ERR "unable to obtain mm\n");
> -               goto failed;
> -       }
> -
> -       /* send remote request */
> -       comm_send_req(p->om.contact, DEP_COMING_HOME);
> -
> -       /* see if other part is with on this */
> -       if (mig_recv_hshake(p->om.contact))
> -               goto failed;
> -
> -       /* receive the process back */
> -       error = mig_do_receive(p);
> -       if (error)
> -               goto failed;
> -
> -       task_clear_dflags(p, DDEPUTY);
> -       link = task_set_comm(p, NULL);
> -       comm_close(link);
> -
> -       return 0;
> -failed:
> -       OMBUG("failed\n");
> -       return -1;
> -}
> -
> /**
>   * task_move_remote2remote - migrate a task from remote to remote
>   * @p:         task to send
> --
> 1.2.4
>
> --
> Ansgar Esztermann
> Researcher & Sysadmin
> http://www2.thphy.uni-duesseldorf.de/~ansgar
>
>
>