[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Cluster-devel] [PATCH] rgmanager: Fix for deadlock



On 09/27/2012 03:18 PM, Ryan McCabe wrote:
> This patch fixes a deadlock in rgmanager that could occur when a node
> starts rgmanager while a service is recovering.


This basically prevents (where other places in the code try to avoid, it
looks like in the comments) cases where RG_START_REMOTE is passed while
a RG_START is already being processed for the same service on the same
node and one or more starts have failed.

ACK.

-- Lon


> 
> Resolves: rhbz#861157
> 
> Signed-off-by: Ryan McCabe <rmccabe redhat com>
> ---
>  rgmanager/src/daemons/rg_state.c   |  1 +
>  rgmanager/src/daemons/rg_thread.c  | 19 ++++++++++++++++++-
>  rgmanager/src/daemons/service_op.c |  1 +
>  3 files changed, 20 insertions(+), 1 deletion(-)
> 
> diff --git a/rgmanager/src/daemons/rg_state.c b/rgmanager/src/daemons/rg_state.c
> index 8c5af5b..80e8667 100644
> --- a/rgmanager/src/daemons/rg_state.c
> +++ b/rgmanager/src/daemons/rg_state.c
> @@ -1963,6 +1963,7 @@ retry:
>  			/* Deliberate */
>  		case RG_EDEPEND:
>  		case RG_EFAIL:
> +		case RG_EDEADLCK:
>  			/* Uh oh - we failed to relocate to this node.
>  			   ensure that we tell the next node to start it from
>  			   the 'recovering' state. */
> diff --git a/rgmanager/src/daemons/rg_thread.c b/rgmanager/src/daemons/rg_thread.c
> index 72b5f96..5e551c3 100644
> --- a/rgmanager/src/daemons/rg_thread.c
> +++ b/rgmanager/src/daemons/rg_thread.c
> @@ -9,6 +9,8 @@
>  #include <rg_queue.h>
>  #include <assert.h>
>  #include <members.h>
> +#include <liblogthread.h>
> +
>  
>  /**
>   * Resource thread list entry.
> @@ -735,13 +737,28 @@ rt_enqueue_request(const char *resgroupname, int request,
>  			ret = 0;
>  			break;
>  		}
> -		fprintf(stderr, "Failed to queue request: Would block\n");
>  		/* EWOULDBLOCK */
>  		pthread_mutex_unlock(resgroup->rt_queue_mutex);
>  		pthread_mutex_unlock(&reslist_mutex);
> +		logt_print(LOG_DEBUG,
> +			"Failed to queue %d request for %s: Would block\n",
> +			request, resgroupname);
>  		return ret;
>  	}
>  
> +	if (resgroup->rt_request == RG_START &&
> +	    (request == RG_START_REMOTE || request == RG_START_RECOVER)) {
> +		send_ret(response_ctx, resgroup->rt_name, RG_EDEADLCK,
> +			request, 0);
> +		msg_free_ctx(response_ctx);
> +		pthread_mutex_unlock(resgroup->rt_queue_mutex);
> +		pthread_mutex_unlock(&reslist_mutex);
> +		logt_print(LOG_DEBUG,
> +			"Failed to queue %d request for %s: Would block\n",
> +			request, resgroupname);
> +		return -1;
> +	}
> +
>  	ret = rq_queue_request(resgroup->rt_queue, resgroup->rt_name,
>  			       request, 0, 0, response_ctx, 0, target,
>  			       arg0, arg1);
> diff --git a/rgmanager/src/daemons/service_op.c b/rgmanager/src/daemons/service_op.c
> index f094129..4b74427 100644
> --- a/rgmanager/src/daemons/service_op.c
> +++ b/rgmanager/src/daemons/service_op.c
> @@ -62,6 +62,7 @@ service_op_start(char *svcName,
>  			++dep;
>  			continue;
>  		case RG_EFAIL:
> +		case RG_EDEADLCK:
>  			++fail;
>  			continue;
>  		case RG_EABORT:
> 


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]