diff options
Diffstat (limited to 'xlators/mgmt/glusterd/src/glusterd-rebalance.c')
| -rw-r--r-- | xlators/mgmt/glusterd/src/glusterd-rebalance.c | 749 |
1 files changed, 749 insertions, 0 deletions
diff --git a/xlators/mgmt/glusterd/src/glusterd-rebalance.c b/xlators/mgmt/glusterd/src/glusterd-rebalance.c new file mode 100644 index 000000000..b7b974c68 --- /dev/null +++ b/xlators/mgmt/glusterd/src/glusterd-rebalance.c @@ -0,0 +1,749 @@ +/* + Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include <inttypes.h> +#include <sys/types.h> +#include <unistd.h> +#include <sys/resource.h> +#include <sys/statvfs.h> + +#include "globals.h" +#include "compat.h" +#include "protocol-common.h" +#include "xlator.h" +#include "logging.h" +#include "timer.h" +#include "glusterd-mem-types.h" +#include "glusterd.h" +#include "glusterd-sm.h" +#include "glusterd-op-sm.h" +#include "glusterd-utils.h" +#include "glusterd-store.h" +#include "run.h" +#include "glusterd-volgen.h" + +#include "syscall.h" +#include "cli1-xdr.h" +#include "xdr-generic.h" + +int32_t +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, glusterd_op_t op) +{ + 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 (this->name, GF_LOG_DEBUG, + "rebalance on volume %s already started", + volinfo->volname); + snprintf (op_errstr, len, "Rebalance on %s is already started", + volinfo->volname); + goto out; + } + + if (glusterd_is_rb_started (volinfo) || + glusterd_is_rb_paused (volinfo)) { + 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 " + "volume %s", volinfo->volname); + goto out; + } + ret = 0; +out: + 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_volinfo_t *volinfo = NULL; + glusterd_defrag_info_t *defrag = NULL; + int ret = 0; + char pidfile[PATH_MAX]; + glusterd_conf_t *priv = NULL; + + priv = THIS->private; + if (!priv) + return 0; + + volinfo = mydata; + if (!volinfo) + return 0; + + defrag = volinfo->rebal.defrag; + if (!defrag) + return 0; + + if ((event == RPC_CLNT_DISCONNECT) && defrag->connected) + volinfo->rebal.defrag = NULL; + + GLUSTERD_GET_DEFRAG_PID_FILE(pidfile, volinfo, priv); + + switch (event) { + case RPC_CLNT_CONNECT: + { + if (defrag->connected) + return 0; + + LOCK (&defrag->lock); + { + defrag->connected = 1; + } + UNLOCK (&defrag->lock); + + gf_log ("", GF_LOG_DEBUG, "%s got RPC_CLNT_CONNECT", + rpc->conn.trans->name); + break; + } + + case RPC_CLNT_DISCONNECT: + { + if (!defrag->connected) + return 0; + + LOCK (&defrag->lock); + { + defrag->connected = 0; + } + UNLOCK (&defrag->lock); + + if (!glusterd_is_service_running (pidfile, NULL)) { + if (volinfo->rebal.defrag_status == + GF_DEFRAG_STATUS_STARTED) { + volinfo->rebal.defrag_status = + GF_DEFRAG_STATUS_FAILED; + } + } + + glusterd_store_perform_node_state_store (volinfo); + + if (defrag->rpc) { + rpc_clnt_unref (defrag->rpc); + defrag->rpc = NULL; + } + if (defrag->cbk_fn) + defrag->cbk_fn (volinfo, + volinfo->rebal.defrag_status); + + GF_FREE (defrag); + gf_log ("", GF_LOG_DEBUG, "%s got RPC_CLNT_DISCONNECT", + rpc->conn.trans->name); + break; + } + default: + gf_log ("", GF_LOG_TRACE, + "got some other RPC event %d", event); + ret = 0; + break; + } + + 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, + glusterd_op_t op) +{ + int ret = -1; + glusterd_defrag_info_t *defrag = NULL; + runner_t runner = {0,}; + glusterd_conf_t *priv = NULL; + char defrag_path[PATH_MAX]; + char sockfile[PATH_MAX] = {0,}; + char pidfile[PATH_MAX] = {0,}; + char logfile[PATH_MAX] = {0,}; + dict_t *options = NULL; + char valgrind_logfile[PATH_MAX] = {0,}; + + priv = THIS->private; + + GF_ASSERT (volinfo); + GF_ASSERT (op_errstr); + + ret = glusterd_defrag_start_validate (volinfo, op_errstr, len, op); + if (ret) + goto out; + if (!volinfo->rebal.defrag) + volinfo->rebal.defrag = + GF_CALLOC (1, sizeof (*volinfo->rebal.defrag), + gf_gld_mt_defrag_info); + if (!volinfo->rebal.defrag) + goto out; + + defrag = volinfo->rebal.defrag; + + defrag->cmd = cmd; + + volinfo->rebal.defrag_cmd = cmd; + volinfo->rebal.op = op; + + LOCK_INIT (&defrag->lock); + + volinfo->rebal.defrag_status = GF_DEFRAG_STATUS_STARTED; + + glusterd_volinfo_reset_defrag_stats (volinfo); + glusterd_store_perform_node_state_store (volinfo); + + GLUSTERD_GET_DEFRAG_DIR (defrag_path, volinfo, priv); + ret = mkdir_p (defrag_path, 0777, _gf_true); + if (ret) { + gf_log (THIS->name, GF_LOG_ERROR, "Failed to create " + "directory %s", defrag_path); + goto out; + } + + GLUSTERD_GET_DEFRAG_SOCK_FILE (sockfile, volinfo, priv); + GLUSTERD_GET_DEFRAG_PID_FILE (pidfile, volinfo, priv); + snprintf (logfile, PATH_MAX, "%s/%s-rebalance.log", + DEFAULT_LOG_FILE_DIRECTORY, volinfo->volname); + runinit (&runner); + + if (priv->valgrind) { + snprintf (valgrind_logfile, PATH_MAX, + "%s/valgrind-%s-rebalance.log", + DEFAULT_LOG_FILE_DIRECTORY, + volinfo->volname); + + runner_add_args (&runner, "valgrind", "--leak-check=full", + "--trace-children=yes", "--track-origins=yes", + NULL); + runner_argprintf (&runner, "--log-file=%s", valgrind_logfile); + } + + runner_add_args (&runner, SBIN_DIR"/glusterfs", + "-s", "localhost", "--volfile-id", volinfo->volname, + "--xlator-option", "*dht.use-readdirp=yes", + "--xlator-option", "*dht.lookup-unhashed=yes", + "--xlator-option", "*dht.assert-no-child-down=yes", + "--xlator-option", "*replicate*.data-self-heal=off", + "--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); + runner_add_arg (&runner, "--xlator-option"); + runner_argprintf (&runner, "*dht.node-uuid=%s", uuid_utoa(MY_UUID)); + runner_add_arg (&runner, "--socket-file"); + runner_argprintf (&runner, "%s",sockfile); + runner_add_arg (&runner, "--pid-file"); + runner_argprintf (&runner, "%s",pidfile); + runner_add_arg (&runner, "-l"); + runner_argprintf (&runner, logfile); + if (volinfo->memory_accounting) + runner_add_arg (&runner, "--mem-accounting"); + + ret = runner_run_nowait (&runner); + if (ret) { + gf_log ("glusterd", GF_LOG_DEBUG, "rebalance command failed"); + goto out; + } + + sleep (5); + + /* Setting frame-timeout to 10mins (600seconds). + * Unix domain sockets ensures that the connection is reliable. The + * default timeout of 30mins used for unreliable network connections is + * too long for unix domain socket connections. + */ + 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; + } + + if (cbk) + defrag->cbk_fn = cbk; + +out: + gf_log ("", GF_LOG_DEBUG, "Returning %d", ret); + return ret; +} + + +int +glusterd_rebalance_rpc_create (glusterd_volinfo_t *volinfo, + glusterd_conf_t *priv, int cmd) +{ + dict_t *options = NULL; + char sockfile[PATH_MAX] = {0,}; + int ret = -1; + glusterd_defrag_info_t *defrag = NULL; + + if (!volinfo->rebal.defrag) + volinfo->rebal.defrag = + GF_CALLOC (1, sizeof (*volinfo->rebal.defrag), + gf_gld_mt_defrag_info); + + if (!volinfo->rebal.defrag) + goto out; + + defrag = volinfo->rebal.defrag; + + defrag->cmd = cmd; + + LOCK_INIT (&defrag->lock); + + GLUSTERD_GET_DEFRAG_SOCK_FILE (sockfile, volinfo, priv); + + /* Setting frame-timeout to 10mins (600seconds). + * Unix domain sockets ensures that the connection is reliable. The + * default timeout of 30mins used for unreliable network connections is + * too long for unix domain socket connections. + */ + 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; + } + ret = 0; +out: + return ret; +} + +int +glusterd_rebalance_cmd_validate (int cmd, char *volname, + glusterd_volinfo_t **volinfo, + char *op_errstr, size_t len) +{ + int ret = -1; + + if (glusterd_volinfo_find(volname, volinfo)) { + gf_log ("glusterd", GF_LOG_ERROR, "Received rebalance on invalid" + " volname %s", volname); + snprintf (op_errstr, len, "Volume %s does not exist", + volname); + goto out; + } + if ((*volinfo)->brick_count <= (*volinfo)->dist_leaf_count) { + gf_log ("glusterd", GF_LOG_ERROR, "Volume %s is not a " + "distribute type or contains only 1 brick", volname); + snprintf (op_errstr, len, "Volume %s is not a distribute " + "volume or contains only 1 brick.\n" + "Not performing rebalance", volname); + goto out; + } + + if ((*volinfo)->status != GLUSTERD_STATUS_STARTED) { + gf_log ("glusterd", GF_LOG_ERROR, "Received rebalance on stopped" + " volname %s", volname); + snprintf (op_errstr, len, "Volume %s needs to " + "be started to perform rebalance", volname); + goto out; + } + + ret = 0; + +out: + gf_log ("glusterd", GF_LOG_DEBUG, "Returning %d", ret); + return ret; +} + +int +__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; + char msg[2048] = {0,}; + xlator_t *this = NULL; + + GF_ASSERT (req); + this = THIS; + GF_ASSERT (this); + + priv = this->private; + GF_ASSERT (priv); + + 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) { + /* Unserialize the dictionary */ + dict = dict_new (); + + ret = dict_unserialize (cli_req.dict.dict_val, + cli_req.dict.dict_len, + &dict); + if (ret < 0) { + 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) { + 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) { + snprintf (msg, sizeof (msg), "Failed to get command"); + gf_log (this->name, GF_LOG_ERROR, "%s", msg); + goto out; + } + + ret = dict_set_static_bin (dict, "node-uuid", MY_UUID, 16); + if (ret) + goto out; + + if ((cmd == GF_DEFRAG_CMD_STATUS) || + (cmd == GF_DEFRAG_CMD_STOP)) { + ret = glusterd_op_begin (req, GD_OP_DEFRAG_BRICK_VOLUME, + dict, msg, sizeof (msg)); + } else + ret = glusterd_op_begin (req, GD_OP_REBALANCE, dict, + msg, sizeof (msg)); + +out: + + glusterd_friend_sm (); + 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, msg); + + } + + free (cli_req.dict.dict_val);//malloced by xdr + + 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) +{ + char *volname = NULL; + int ret = 0; + int32_t cmd = 0; + char msg[2048] = {0}; + glusterd_volinfo_t *volinfo = NULL; + char *task_id_str = NULL; + dict_t *op_ctx = NULL; + xlator_t *this = 0; + + this = THIS; + GF_ASSERT (this); + + ret = dict_get_str (dict, "volname", &volname); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "volname not found"); + goto out; + } + + ret = dict_get_int32 (dict, "rebalance-command", &cmd); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "cmd not found"); + goto out; + } + + ret = glusterd_rebalance_cmd_validate (cmd, volname, &volinfo, + msg, sizeof (msg)); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "failed to validate"); + goto out; + } + switch (cmd) { + case GF_DEFRAG_CMD_START: + case GF_DEFRAG_CMD_START_LAYOUT_FIX: + case GF_DEFRAG_CMD_START_FORCE: + if (is_origin_glusterd (dict)) { + op_ctx = glusterd_op_get_ctx (); + if (!op_ctx) { + ret = -1; + gf_log (this->name, GF_LOG_ERROR, + "Failed to get op_ctx"); + goto out; + } + + ret = glusterd_generate_and_set_task_id + (op_ctx, GF_REBALANCE_TID_KEY); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "Failed to generate task-id"); + goto out; + } + } else { + ret = dict_get_str (dict, GF_REBALANCE_TID_KEY, + &task_id_str); + if (ret) { + snprintf (msg, sizeof (msg), + "Missing rebalance-id"); + gf_log (this->name, GF_LOG_WARNING, "%s", msg); + ret = 0; + } + } + ret = glusterd_defrag_start_validate (volinfo, msg, + sizeof (msg), + GD_OP_REBALANCE); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, + "start validate failed"); + goto out; + } + break; + case GF_DEFRAG_CMD_STATUS: + case GF_DEFRAG_CMD_STOP: + break; + default: + break; + } + + ret = 0; +out: + if (ret && op_errstr && msg[0]) + *op_errstr = gf_strdup (msg); + + return ret; +} + + +int +glusterd_op_rebalance (dict_t *dict, char **op_errstr, dict_t *rsp_dict) +{ + char *volname = NULL; + int ret = 0; + int32_t cmd = 0; + char msg[2048] = {0}; + glusterd_volinfo_t *volinfo = NULL; + glusterd_conf_t *priv = NULL; + glusterd_brickinfo_t *brickinfo = NULL; + glusterd_brickinfo_t *tmp = NULL; + gf_boolean_t volfile_update = _gf_false; + char *task_id_str = NULL; + dict_t *ctx = NULL; + xlator_t *this = NULL; + + this = THIS; + GF_ASSERT (this); + priv = this->private; + + ret = dict_get_str (dict, "volname", &volname); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "volname not given"); + goto out; + } + + ret = dict_get_int32 (dict, "rebalance-command", &cmd); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "command not given"); + goto out; + } + + + ret = glusterd_rebalance_cmd_validate (cmd, volname, &volinfo, + msg, sizeof (msg)); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "cmd validate failed"); + goto out; + } + + /* Set task-id, if available, in op_ctx dict for operations other than + * start + */ + if (cmd == GF_DEFRAG_CMD_STATUS || cmd == GF_DEFRAG_CMD_STOP) { + if (!uuid_is_null (volinfo->rebal.rebalance_id)) { + ctx = glusterd_op_get_ctx (); + if (!ctx) { + gf_log (this->name, GF_LOG_ERROR, + "Failed to get op_ctx"); + ret = -1; + goto out; + } + + if (GD_OP_REMOVE_BRICK == volinfo->rebal.op) + ret = glusterd_copy_uuid_to_dict + (volinfo->rebal.rebalance_id, ctx, + GF_REMOVE_BRICK_TID_KEY); + else + ret = glusterd_copy_uuid_to_dict + (volinfo->rebal.rebalance_id, ctx, + GF_REBALANCE_TID_KEY); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "Failed to set task-id"); + goto out; + } + } + } + + switch (cmd) { + case GF_DEFRAG_CMD_START: + case GF_DEFRAG_CMD_START_LAYOUT_FIX: + case GF_DEFRAG_CMD_START_FORCE: + ret = dict_get_str (dict, GF_REBALANCE_TID_KEY, &task_id_str); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "Missing rebalance " + "id"); + 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) { + if (!brickinfo->decommissioned) + continue; + brickinfo->decommissioned = 0; + volfile_update = _gf_true; + } + + if (volfile_update == _gf_false) { + ret = 0; + break; + } + + ret = glusterd_create_volfiles_and_notify_services (volinfo); + if (ret) { + gf_log (this->name, GF_LOG_WARNING, + "failed to create volfiles"); + goto out; + } + + ret = glusterd_store_volinfo (volinfo, + GLUSTERD_VOLINFO_VER_AC_INCREMENT); + if (ret) { + gf_log (this->name, GF_LOG_WARNING, + "failed to store volinfo"); + goto out; + } + + ret = 0; + break; + + case GF_DEFRAG_CMD_STATUS: + break; + default: + break; + } + +out: + if (ret && op_errstr && msg[0]) + *op_errstr = gf_strdup (msg); + + return ret; +} + +int32_t +glusterd_defrag_event_notify_handle (dict_t *dict) +{ + glusterd_volinfo_t *volinfo = NULL; + char *volname = NULL; + int32_t ret = -1; + + ret = dict_get_str (dict, "volname", &volname); + if (ret) { + gf_log ("", GF_LOG_ERROR, "Failed to get volname"); + return ret; + } + + ret = glusterd_volinfo_find (volname, &volinfo); + if (ret) { + gf_log ("", GF_LOG_ERROR, "Failed to get volinfo for %s" + , volname); + return ret; + } + + ret = glusterd_defrag_volume_status_update (volinfo, dict); + + if (ret) + gf_log ("", GF_LOG_ERROR, "Failed to update status"); + return ret; +} |
