vine: vine_temp.c

doc/manuals/taskvine/index.md

@@ -2750,6 +2750,7 @@ change.
 |-----------|-------------|---------------|
 | attempt-schedule-depth | The amount of tasks to attempt scheduling on each pass of send_one_task in the main loop. | 100 |
 | category-steady-n-tasks | Minimum number of successful tasks to use a sample for automatic resource allocation modes after encountering a new resource maximum. | 25 |
+| clean-redundant-replicas | Remove redundant temporary file replicas to save worker's local disk space. | 0 |
 | default-transfer-rate | The assumed network bandwidth used until sufficient data has been collected.  (1MB/s)
 | disconnect-slow-workers-factor | Set the multiplier of the average task time at which point to disconnect a worker; disabled if less than 1. (default=0)
 | hungry-minimum          | Smallest number of waiting tasks in the manager before declaring it hungry | 10 |
@@ -2767,6 +2768,7 @@ change.
 | ramp-down-heuristic     | If set to 1 and there are more workers than tasks waiting, then tasks are allocated all the free resources of a worker large enough to run them. If monitoring watchdog is not enabled, then this heuristic has no effect. | 0 |
 | resource-submit-multiplier | Assume that workers have `resource x resources-submit-multiplier` available.<br> This overcommits resources at the worker, causing tasks to be sent to workers that cannot be immediately executed.<br>The extra tasks wait at the worker until resources become available. | 1 |
 | sandbox-grow-factor    | When task disk sandboxes are exhausted, increase the allocation using their measured valued times this factor. Minimum is 1.1. | 2 |
+| shift-disk-load | Proactively shift temporary files away from the most disk-heavy worker to those with more available disk. | 0 |
 | short-timeout | Set the minimum timeout in seconds when sending a brief message to a single worker. | 5 |
 | temp-replica-count    | Number of temp file replicas created across workers | 0 |
 | transfer-outlier-factor | Transfer that are this many times slower than the average will be terminated. | 10 |

taskvine/src/manager/Makefile

@@ -28,7 +28,8 @@ SOURCES = \
 	vine_file_replica_table.c \
 	vine_fair.c \
 	vine_runtime_dir.c \
-	vine_task_groups.c
+	vine_task_groups.c \
+	vine_temp.c
 
 PUBLIC_HEADERS = taskvine.h
 

taskvine/src/manager/vine_file_replica_table.c

@@ -147,94 +147,6 @@ struct vine_worker_info *vine_file_replica_table_find_worker(struct vine_manager
 	return peer_selected;
 }
 
-// trigger replications of file to satisfy temp_replica_count
-int vine_file_replica_table_replicate(struct vine_manager *m, struct vine_file *f, struct set *source_workers, int to_find)
-{
-	if (!f || !source_workers) {
-		return 0;
-	}
-
-	int nsource_workers = set_size(source_workers);
-	int round_replication_request_sent = 0;
-
-	/* get the elements of set so we can insert new replicas to sources */
-	struct vine_worker_info **source_workers_frozen = (struct vine_worker_info **)set_values(source_workers);
-	struct vine_worker_info *source_worker;
-
-	for (int i = 0; i < nsource_workers; i++) {
-
-		source_worker = source_workers_frozen[i];
-		int dest_workers_found = 0;
-
-		// skip if the file on the source is not ready to transfer
-		struct vine_file_replica *replica = hash_table_lookup(source_worker->current_files, f->cached_name);
-		if (!replica || replica->state != VINE_FILE_REPLICA_STATE_READY) {
-			continue;
-		}
-
-		// skip if the source is busy with other transfers
-		if (source_worker->outgoing_xfer_counter >= m->worker_source_max_transfers) {
-			continue;
-		}
-
-		char *source_addr = string_format("%s/%s", source_worker->transfer_url, f->cached_name);
-
-		char *id;
-		struct vine_worker_info *dest_worker;
-		int offset_bookkeep;
-
-		HASH_TABLE_ITERATE_RANDOM_START(m->worker_table, offset_bookkeep, id, dest_worker)
-		{
-			// skip if the source and destination are on the same host
-			if (set_lookup(source_workers, dest_worker) || strcmp(source_worker->hostname, dest_worker->hostname) == 0) {
-				continue;
-			}
-
-			// skip if the destination is not ready to transfer
-			if (!dest_worker->transfer_port_active) {
-				continue;
-			}
-
-			// skip if the destination is draining
-			if (dest_worker->draining) {
-				continue;
-			}
-
-			// skip if the destination is busy with other transfers
-			if (dest_worker->incoming_xfer_counter >= m->worker_source_max_transfers) {
-				continue;
-			}
-
-			debug(D_VINE, "replicating %s from %s to %s", f->cached_name, source_worker->addrport, dest_worker->addrport);
-
-			vine_manager_put_url_now(m, dest_worker, source_worker, source_addr, f);
-
-			round_replication_request_sent++;
-
-			// break if we have found enough destinations for this source
-			if (++dest_workers_found >= MIN(m->file_source_max_transfers, to_find)) {
-				break;
-			}
-
-			// break if the source becomes busy with transfers
-			if (source_worker->outgoing_xfer_counter >= m->worker_source_max_transfers) {
-				break;
-			}
-		}
-
-		free(source_addr);
-
-		// break if we have sent enough replication requests for this file
-		if (round_replication_request_sent >= to_find) {
-			break;
-		}
-	}
-
-	free(source_workers_frozen);
-
-	return round_replication_request_sent;
-}
-
 /*
 Count number of replicas of a file in the system.
 */
@@ -258,32 +170,6 @@ int vine_file_replica_table_count_replicas(struct vine_manager *q, const char *c
 	return count;
 }
 
-/*
-Check if a file replica exists on a worker. We accept both CREATING and READY replicas,
-since a CREATING replica may already exist physically but hasn't yet received the cache-update
-message from the manager. However, we do not accept DELETING replicas, as they indicate
-the source worker has already been sent an unlink request—any subsequent cache-update or
-cache-invalid events will lead to deletion.
-*/
-int vine_file_replica_table_exists_somewhere(struct vine_manager *q, const char *cachename)
-{
-	struct set *workers = hash_table_lookup(q->file_worker_table, cachename);
-	if (!workers || set_size(workers) < 1) {
-		return 0;
-	}
-
-	struct vine_worker_info *w;
-	SET_ITERATE(workers, w)
-	{
-		struct vine_file_replica *r = vine_file_replica_table_lookup(w, cachename);
-		if (r && (r->state == VINE_FILE_REPLICA_STATE_CREATING || r->state == VINE_FILE_REPLICA_STATE_READY)) {
-			return 1;
-		}
-	}
-
-	return 0;
-}
-
 // get or create a replica for a worker and cachename
 struct vine_file_replica *vine_file_replica_table_get_or_create(struct vine_manager *m, struct vine_worker_info *w, const char *cachename, vine_file_type_t type, vine_cache_level_t cache_level, int64_t size, time_t mtime)
 {

taskvine/src/manager/vine_file_replica_table.h

@@ -30,10 +30,6 @@ struct vine_file_replica *vine_file_replica_table_get_or_create(struct vine_mana
 
 struct vine_worker_info *vine_file_replica_table_find_worker(struct vine_manager *q, const char *cachename);
 
-int vine_file_replica_table_replicate(struct vine_manager *q, struct vine_file *f, struct set *source_workers, int to_find);
-
-int vine_file_replica_table_exists_somewhere( struct vine_manager *q, const char *cachename );
-
 int vine_file_replica_table_count_replicas( struct vine_manager *q, const char *cachename, vine_file_replica_state_t state );
 
 #endif