[lttng-dev] [PATCH 2/3] rculfhash: handle pthread_create failures

[Eric Wong]

Like calloc, pthread_create may fail with EAGAIN due to a lack
of resources.  Account for that and gracefully continue.

Signed-off-by: Eric Wong <normalperson at yhbt.net>
---
 rculfhash.c | 24 ++++++++++++++++++++++--
 1 file changed, 22 insertions(+), 2 deletions(-)

diff --git a/rculfhash.c b/rculfhash.c
index 2a45045..d534be2 100644
--- a/rculfhash.c
+++ b/rculfhash.c
@@ -1171,6 +1171,7 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i,
 		void (*fct)(struct cds_lfht *ht, unsigned long i,
 			unsigned long start, unsigned long len))
 {
+	unsigned long start = 0;
 	unsigned long partition_len;
 	struct partition_resize_work *work;
 	int thread, ret;
@@ -1201,6 +1202,18 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i,
 		work[thread].fct = fct;
 		ret = pthread_create(&(work[thread].thread_id), ht->resize_attr,
 			partition_resize_thread, &work[thread]);
+		if (ret == EAGAIN) {
+			/*
+			 * out of resources: wait and join the threads
+			 * we've created, then handle leftovers
+			 */
+			dbg_printf(
+			      "error spawning for resize, single-threading\n");
+			start = work[thread].start;
+			len -= start;
+			nr_threads = thread;
+			break;
+		}
 		assert(!ret);
 	}
 	for (thread = 0; thread < nr_threads; thread++) {
@@ -1208,10 +1221,17 @@ void partition_resize_helper(struct cds_lfht *ht, unsigned long i,
 		assert(!ret);
 	}
 	free(work);
-	return;
+
+	/*
+	 * a pthread_create failure above will either lead in us having
+	 * no threads to join or starting at a non-zero offset,
+	 * fallback to single thread processing of leftovers
+	 */
+	if (start == 0 && nr_threads > 0)
+		return;
 fallback:
 	ht->flavor->thread_online();
-	fct(ht, i, 0, len);
+	fct(ht, i, start, len);
 	ht->flavor->thread_offline();
 }
 
-- 
2.0.0.259.gbf1bc9c