wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-08-20 12:40:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix local scheduler crash when driver creates actor and exits. (#1474)

Browse Source 
* Make check failures in redis.cc more informative.

* Fix bug by calling task_table_add_task.

* Add test.
This commit is contained in:
Robert Nishihara
2018-01-26 14:29:53 -08:00
committed by Philipp Moritz
parent 668737f383
commit 3195c6aa63
3 changed files with 75 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/common/state/redis.cc
+11 -11
View File
@@ -225,7 +225,7 @@ void db_connect_shard(const std::string &db_address,
* we've defined. */
reply = (redisReply *) redisCommandArgv(sync_context, argc, argv, argvlen);
CHECKM(reply != NULL, "db_connect failed on RAY.CONNECT");
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
freeReplyObject(reply);
free(argv);
@@ -326,7 +326,7 @@ void db_disconnect(DBHandle *db) {
redisReply *reply =
(redisReply *) redisCommand(db->sync_context, "RAY.DISCONNECT %b",
db->client.data(), sizeof(db->client));
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
freeReplyObject(reply);
@@ -382,7 +382,7 @@ void redis_object_table_add_callback(redisAsyncContext *c,
"because a nondeterministic task was executed twice, either for "
"reconstruction or for speculation.");
} else {
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
}
/* Call the done callback if there is one. */
@@ -428,7 +428,7 @@ void redis_object_table_remove_callback(redisAsyncContext *c,
* condition with an object_table_add. */
return;
}
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
/* Call the done callback if there is one. */
if (callback_data->done_callback != NULL) {
@@ -486,7 +486,7 @@ void redis_result_table_add_callback(redisAsyncContext *c,
REDIS_CALLBACK_HEADER(db, callback_data, r);
redisReply *reply = (redisReply *) r;
/* Check that the command succeeded. */
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strncmp(reply->str, "OK", strlen("OK")) == 0, "reply->str is %s",
reply->str);
/* Call the done callback if there is one. */
@@ -802,7 +802,7 @@ void redis_object_table_request_notifications_callback(redisAsyncContext *c,
/* Do some minimal checking. */
redisReply *reply = (redisReply *) r;
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
CHECK(callback_data->done_callback == NULL);
/* Clean up the timer and callback. */
@@ -909,7 +909,7 @@ void redis_task_table_add_task_callback(redisAsyncContext *c,
callback_data->data->Get());
}
} else {
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
/* Call the done callback if there is one. */
if (callback_data->done_callback != NULL) {
@@ -974,7 +974,7 @@ void redis_task_table_update_callback(redisAsyncContext *c,
callback_data->data->Get());
}
} else {
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
/* Call the done callback if there is one. */
@@ -1194,7 +1194,7 @@ void redis_db_client_table_remove_callback(redisAsyncContext *c,
REDIS_CALLBACK_HEADER(db, callback_data, r);
redisReply *reply = (redisReply *) r;
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
/* Call the done callback if there is one. */
@@ -1424,7 +1424,7 @@ void redis_local_scheduler_table_disconnect(DBHandle *db) {
redisReply *reply = (redisReply *) redisCommand(
db->sync_context, "PUBLISH local_schedulers %b", fbb.GetBufferPointer(),
(size_t) fbb.GetSize());
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECK(reply->type == REDIS_REPLY_INTEGER);
LOG_DEBUG("%lld subscribers received this publish.\n", reply->integer);
freeReplyObject(reply);
@@ -1632,7 +1632,7 @@ void redis_push_error_hmset_callback(redisAsyncContext *c,
redisReply *reply = (redisReply *) r;
/* Make sure we were able to add the error information. */
CHECK(reply->type != REDIS_REPLY_ERROR);
CHECKM(reply->type != REDIS_REPLY_ERROR, "reply->str is %s", reply->str);
CHECKM(strcmp(reply->str, "OK") == 0, "reply->str is %s", reply->str);
/* Add the error to this driver's list of errors. */
src/local_scheduler/local_scheduler_algorithm.cc
+5 -1
View File
@@ -407,7 +407,11 @@ void finish_killed_task(LocalSchedulerState *state,
if (state->db != NULL) {
Task *task = Task_alloc(execution_spec, TASK_STATUS_DONE,
get_db_client_id(state->db));
task_table_update(state->db, task, NULL, NULL, NULL);
// In most cases, task_table_update would be appropriate, however, it is
// possible in some cases that the task has not yet been added to the task
// table (e.g., if it is an actor task that is queued locally because the
// actor has not been created yet).
task_table_add_task(state->db, task, NULL, NULL, NULL);
}
}