summaryrefslogtreecommitdiffstats
path: root/xlators/mgmt/glusterd/src/glusterd-rebalance.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/mgmt/glusterd/src/glusterd-rebalance.c')
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-rebalance.c136
1 files changed, 93 insertions, 43 deletions
diff --git a/xlators/mgmt/glusterd/src/glusterd-rebalance.c b/xlators/mgmt/glusterd/src/glusterd-rebalance.c
index 4c7282e82..b7b974c68 100644
--- a/xlators/mgmt/glusterd/src/glusterd-rebalance.c
+++ b/xlators/mgmt/glusterd/src/glusterd-rebalance.c
@@ -42,12 +42,27 @@ glusterd_brick_op_cbk (struct rpc_req *req, struct iovec *iov,
int count, void *myframe);
int
glusterd_defrag_start_validate (glusterd_volinfo_t *volinfo, char *op_errstr,
- size_t len)
+ size_t len, glusterd_op_t op)
{
- int ret = -1;
+ int ret = -1;
+ xlator_t *this = NULL;
+
+ this = THIS;
+ GF_ASSERT (this);
+
+ /* Check only if operation is not remove-brick */
+ if ((GD_OP_REMOVE_BRICK != op) &&
+ !gd_is_remove_brick_committed (volinfo)) {
+ gf_log (this->name, GF_LOG_DEBUG, "A remove-brick task on "
+ "volume %s is not yet committed", volinfo->volname);
+ snprintf (op_errstr, len, "A remove-brick task on volume %s is"
+ " not yet committed. Either commit or stop the "
+ "remove-brick task.", volinfo->volname);
+ goto out;
+ }
if (glusterd_is_defrag_on (volinfo)) {
- gf_log ("glusterd", GF_LOG_DEBUG,
+ gf_log (this->name, GF_LOG_DEBUG,
"rebalance on volume %s already started",
volinfo->volname);
snprintf (op_errstr, len, "Rebalance on %s is already started",
@@ -57,7 +72,7 @@ glusterd_defrag_start_validate (glusterd_volinfo_t *volinfo, char *op_errstr,
if (glusterd_is_rb_started (volinfo) ||
glusterd_is_rb_paused (volinfo)) {
- gf_log ("glusterd", GF_LOG_DEBUG,
+ gf_log (this->name, GF_LOG_DEBUG,
"Rebalance failed as replace brick is in progress on volume %s",
volinfo->volname);
snprintf (op_errstr, len, "Rebalance failed as replace brick is in progress on "
@@ -66,13 +81,14 @@ glusterd_defrag_start_validate (glusterd_volinfo_t *volinfo, char *op_errstr,
}
ret = 0;
out:
- gf_log ("glusterd", GF_LOG_DEBUG, "Returning %d", ret);
+ gf_log (this->name, GF_LOG_DEBUG, "Returning %d", ret);
return ret;
}
+
int32_t
-glusterd_defrag_notify (struct rpc_clnt *rpc, void *mydata,
- rpc_clnt_event_t event, void *data)
+__glusterd_defrag_notify (struct rpc_clnt *rpc, void *mydata,
+ rpc_clnt_event_t event, void *data)
{
glusterd_volinfo_t *volinfo = NULL;
glusterd_defrag_info_t *defrag = NULL;
@@ -130,8 +146,6 @@ glusterd_defrag_notify (struct rpc_clnt *rpc, void *mydata,
GF_DEFRAG_STATUS_STARTED) {
volinfo->rebal.defrag_status =
GF_DEFRAG_STATUS_FAILED;
- } else {
- volinfo->rebal.defrag_cmd = 0;
}
}
@@ -160,6 +174,14 @@ glusterd_defrag_notify (struct rpc_clnt *rpc, void *mydata,
return ret;
}
+int32_t
+glusterd_defrag_notify (struct rpc_clnt *rpc, void *mydata,
+ rpc_clnt_event_t event, void *data)
+{
+ return glusterd_big_locked_notify (rpc, mydata, event,
+ data, __glusterd_defrag_notify);
+}
+
int
glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
size_t len, int cmd, defrag_cbk_fn_t cbk,
@@ -174,15 +196,14 @@ glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
char pidfile[PATH_MAX] = {0,};
char logfile[PATH_MAX] = {0,};
dict_t *options = NULL;
-#ifdef DEBUG
char valgrind_logfile[PATH_MAX] = {0,};
-#endif
+
priv = THIS->private;
GF_ASSERT (volinfo);
GF_ASSERT (op_errstr);
- ret = glusterd_defrag_start_validate (volinfo, op_errstr, len);
+ ret = glusterd_defrag_start_validate (volinfo, op_errstr, len, op);
if (ret)
goto out;
if (!volinfo->rebal.defrag)
@@ -196,6 +217,7 @@ glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
defrag->cmd = cmd;
+ volinfo->rebal.defrag_cmd = cmd;
volinfo->rebal.op = op;
LOCK_INIT (&defrag->lock);
@@ -218,7 +240,7 @@ glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
snprintf (logfile, PATH_MAX, "%s/%s-rebalance.log",
DEFAULT_LOG_FILE_DIRECTORY, volinfo->volname);
runinit (&runner);
-#ifdef DEBUG
+
if (priv->valgrind) {
snprintf (valgrind_logfile, PATH_MAX,
"%s/valgrind-%s-rebalance.log",
@@ -226,10 +248,10 @@ glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
volinfo->volname);
runner_add_args (&runner, "valgrind", "--leak-check=full",
- "--trace-children=yes", NULL);
+ "--trace-children=yes", "--track-origins=yes",
+ NULL);
runner_argprintf (&runner, "--log-file=%s", valgrind_logfile);
}
-#endif
runner_add_args (&runner, SBIN_DIR"/glusterfs",
"-s", "localhost", "--volfile-id", volinfo->volname,
@@ -240,6 +262,8 @@ glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
"--xlator-option",
"*replicate*.metadata-self-heal=off",
"--xlator-option", "*replicate*.entry-self-heal=off",
+ "--xlator-option", "*replicate*.readdir-failover=off",
+ "--xlator-option", "*dht.readdir-optimize=on",
NULL);
runner_add_arg (&runner, "--xlator-option");
runner_argprintf ( &runner, "*dht.rebalance-cmd=%d",cmd);
@@ -267,14 +291,16 @@ glusterd_handle_defrag_start (glusterd_volinfo_t *volinfo, char *op_errstr,
* default timeout of 30mins used for unreliable network connections is
* too long for unix domain socket connections.
*/
- ret = rpc_clnt_transport_unix_options_build (&options, sockfile, 600);
+ ret = rpc_transport_unix_options_build (&options, sockfile, 600);
if (ret) {
gf_log (THIS->name, GF_LOG_ERROR, "Unix options build failed");
goto out;
}
+ synclock_unlock (&priv->big_lock);
ret = glusterd_rpc_create (&defrag->rpc, options,
glusterd_defrag_notify, volinfo);
+ synclock_lock (&priv->big_lock);
if (ret) {
gf_log (THIS->name, GF_LOG_ERROR, "RPC create failed");
goto out;
@@ -319,14 +345,16 @@ glusterd_rebalance_rpc_create (glusterd_volinfo_t *volinfo,
* default timeout of 30mins used for unreliable network connections is
* too long for unix domain socket connections.
*/
- ret = rpc_clnt_transport_unix_options_build (&options, sockfile, 600);
+ ret = rpc_transport_unix_options_build (&options, sockfile, 600);
if (ret) {
gf_log (THIS->name, GF_LOG_ERROR, "Unix options build failed");
goto out;
}
+ synclock_unlock (&priv->big_lock);
ret = glusterd_rpc_create (&defrag->rpc, options,
glusterd_defrag_notify, volinfo);
+ synclock_lock (&priv->big_lock);
if (ret) {
gf_log (THIS->name, GF_LOG_ERROR, "RPC create failed");
goto out;
@@ -375,26 +403,32 @@ out:
}
int
-glusterd_handle_defrag_volume (rpcsvc_request_t *req)
+__glusterd_handle_defrag_volume (rpcsvc_request_t *req)
{
- int32_t ret = -1;
- gf_cli_req cli_req = {{0,}};
- glusterd_conf_t *priv = NULL;
- dict_t *dict = NULL;
- char *volname = NULL;
- gf_cli_defrag_type cmd = 0;
+ int32_t ret = -1;
+ gf_cli_req cli_req = {{0,}};
+ glusterd_conf_t *priv = NULL;
+ dict_t *dict = NULL;
+ char *volname = NULL;
+ gf_cli_defrag_type cmd = 0;
+ char msg[2048] = {0,};
+ xlator_t *this = NULL;
GF_ASSERT (req);
+ this = THIS;
+ GF_ASSERT (this);
- priv = THIS->private;
+ priv = this->private;
+ GF_ASSERT (priv);
- if (!xdr_to_generic (req->msg[0], &cli_req,
- (xdrproc_t)xdr_gf_cli_req)) {
+ ret = xdr_to_generic (req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+ if (ret < 0) {
//failed to decode msg;
req->rpc_err = GARBAGE_ARGS;
goto out;
}
- if (cli_req.dict.dict_len) {
+
+ if (cli_req.dict.dict_len) {
/* Unserialize the dictionary */
dict = dict_new ();
@@ -402,24 +436,25 @@ glusterd_handle_defrag_volume (rpcsvc_request_t *req)
cli_req.dict.dict_len,
&dict);
if (ret < 0) {
- gf_log ("glusterd", GF_LOG_ERROR,
- "failed to "
+ gf_log (this->name, GF_LOG_ERROR, "failed to "
"unserialize req-buffer to dictionary");
+ snprintf (msg, sizeof (msg), "Unable to decode the "
+ "command");
goto out;
}
}
ret = dict_get_str (dict, "volname", &volname);
if (ret) {
- gf_log (THIS->name, GF_LOG_ERROR,
- "Failed to get volname");
+ snprintf (msg, sizeof (msg), "Failed to get volume name");
+ gf_log (this->name, GF_LOG_ERROR, "%s", msg);
goto out;
}
ret = dict_get_int32 (dict, "rebalance-command", (int32_t*)&cmd);
if (ret) {
- gf_log (THIS->name, GF_LOG_ERROR,
- "Failed to get command");
+ snprintf (msg, sizeof (msg), "Failed to get command");
+ gf_log (this->name, GF_LOG_ERROR, "%s", msg);
goto out;
}
@@ -430,9 +465,10 @@ glusterd_handle_defrag_volume (rpcsvc_request_t *req)
if ((cmd == GF_DEFRAG_CMD_STATUS) ||
(cmd == GF_DEFRAG_CMD_STOP)) {
ret = glusterd_op_begin (req, GD_OP_DEFRAG_BRICK_VOLUME,
- dict);
+ dict, msg, sizeof (msg));
} else
- ret = glusterd_op_begin (req, GD_OP_REBALANCE, dict);
+ ret = glusterd_op_begin (req, GD_OP_REBALANCE, dict,
+ msg, sizeof (msg));
out:
@@ -440,11 +476,10 @@ out:
glusterd_op_sm ();
if (ret) {
+ if (msg[0] == '\0')
+ snprintf (msg, sizeof (msg), "Operation failed");
ret = glusterd_op_send_cli_response (GD_OP_REBALANCE, ret, 0,
- req, dict,
- "operation failed");
- if (dict)
- dict_unref (dict);
+ req, dict, msg);
}
@@ -453,6 +488,12 @@ out:
return 0;
}
+int
+glusterd_handle_defrag_volume (rpcsvc_request_t *req)
+{
+ return glusterd_big_locked_handler (req, __glusterd_handle_defrag_volume);
+}
+
int
glusterd_op_stage_rebalance (dict_t *dict, char **op_errstr)
@@ -491,7 +532,7 @@ glusterd_op_stage_rebalance (dict_t *dict, char **op_errstr)
case GF_DEFRAG_CMD_START:
case GF_DEFRAG_CMD_START_LAYOUT_FIX:
case GF_DEFRAG_CMD_START_FORCE:
- if (is_origin_glusterd ()) {
+ if (is_origin_glusterd (dict)) {
op_ctx = glusterd_op_get_ctx ();
if (!op_ctx) {
ret = -1;
@@ -517,8 +558,9 @@ glusterd_op_stage_rebalance (dict_t *dict, char **op_errstr)
ret = 0;
}
}
- ret = glusterd_defrag_start_validate (volinfo,
- msg, sizeof (msg));
+ ret = glusterd_defrag_start_validate (volinfo, msg,
+ sizeof (msg),
+ GD_OP_REBALANCE);
if (ret) {
gf_log (this->name, GF_LOG_DEBUG,
"start validate failed");
@@ -621,11 +663,19 @@ glusterd_op_rebalance (dict_t *dict, char **op_errstr, dict_t *rsp_dict)
ret = 0;
} else {
uuid_parse (task_id_str, volinfo->rebal.rebalance_id) ;
+ volinfo->rebal.op = GD_OP_REBALANCE;
}
ret = glusterd_handle_defrag_start (volinfo, msg, sizeof (msg),
cmd, NULL, GD_OP_REBALANCE);
break;
case GF_DEFRAG_CMD_STOP:
+ /* Clear task-id only on explicitly stopping rebalance.
+ * Also clear the stored operation, so it doesn't cause trouble
+ * with future rebalance/remove-brick starts
+ */
+ uuid_clear (volinfo->rebal.rebalance_id);
+ volinfo->rebal.op = GD_OP_NONE;
+
/* Fall back to the old volume file in case of decommission*/
list_for_each_entry_safe (brickinfo, tmp, &volinfo->bricks,
brick_list) {
generated by cgit (git 2.34.1) at 2026-04-12 06:33:26 +0000