diff options
author | Kaushal M <kaushal@redhat.com> | 2012-09-14 11:45:34 +0530 |
---|---|---|
committer | Anand Avati <avati@redhat.com> | 2012-12-19 13:32:49 -0800 |
commit | 5eb8bac561b7374589bd72d597ed7eec95aa7de6 (patch) | |
tree | 028a5aa6582b0497cf0ffda62bf023bc88b391b3 /xlators/mgmt/glusterd/src/glusterd-replace-brick.c | |
parent | da7ca1efcf3a621c27f05d621715e57fdc5aa397 (diff) |
glusterd, cli: Task id's for async tasks
This patch introduces task-id's for async tasks like rebalance, remove-brick and
replace-brick. An id is generated for each task when it is started and displayed
to the user in cli output. The status of running tasks is also included in the
output of "volume status" along with its id, so that a user can easily track the
progress of an async task.
Also,
* added tests for this feature into the regression test suite.
* added a python script for creating files, 'create-files.py', courtesy
Vijaykumar Koppad (vkoppad@redhat.com) into the test suite.
This patch reverts the revert commit 698deb33d731df6de84da8ae8ee4045e1543a168.
BUG: 857330
Change-Id: Id43d7cb629a38f47f733fbc18cb4c5f2f0327c7a
Signed-off-by: Kaushal M <kaushal@redhat.com>
Reviewed-on: http://review.gluster.org/4294
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
Diffstat (limited to 'xlators/mgmt/glusterd/src/glusterd-replace-brick.c')
-rw-r--r-- | xlators/mgmt/glusterd/src/glusterd-replace-brick.c | 219 |
1 files changed, 145 insertions, 74 deletions
diff --git a/xlators/mgmt/glusterd/src/glusterd-replace-brick.c b/xlators/mgmt/glusterd/src/glusterd-replace-brick.c index 73e94781717..94f498d33bc 100644 --- a/xlators/mgmt/glusterd/src/glusterd-replace-brick.c +++ b/xlators/mgmt/glusterd/src/glusterd-replace-brick.c @@ -180,7 +180,7 @@ glusterd_get_rb_dst_brickinfo (glusterd_volinfo_t *volinfo, if (!volinfo || !brickinfo) goto out; - *brickinfo = volinfo->dst_brick; + *brickinfo = volinfo->rep_brick.dst_brick; ret = 0; @@ -211,38 +211,43 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, char *savetok = NULL; char voldir[PATH_MAX] = {0}; char pidfile[PATH_MAX] = {0}; + char *task_id_str = NULL; + xlator_t *this = NULL; - priv = THIS->private; + this = THIS; + GF_ASSERT (this); + + priv = this->private; GF_ASSERT (priv); ret = dict_get_str (dict, "src-brick", &src_brick); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get src brick"); + gf_log (this->name, GF_LOG_ERROR, "Unable to get src brick"); goto out; } - gf_log ("", GF_LOG_DEBUG, "src brick=%s", src_brick); + gf_log (this->name, GF_LOG_DEBUG, "src brick=%s", src_brick); ret = dict_get_str (dict, "dst-brick", &dst_brick); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get dest brick"); + gf_log (this->name, GF_LOG_ERROR, "Unable to get dest brick"); goto out; } - gf_log ("", GF_LOG_DEBUG, "dst brick=%s", dst_brick); + gf_log (this->name, GF_LOG_DEBUG, "dst brick=%s", dst_brick); ret = dict_get_str (dict, "volname", &volname); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get volume name"); + gf_log (this->name, GF_LOG_ERROR, "Unable to get volume name"); goto out; } ret = dict_get_int32 (dict, "operation", (int32_t *)&replace_op); if (ret) { - gf_log ("", GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_DEBUG, "dict get on replace-brick operation failed"); goto out; } @@ -274,7 +279,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, !glusterd_is_valid_volfpath (volname, dst_brick)) { snprintf (msg, sizeof (msg), "brick path %s is too " "long.", dst_brick); - gf_log ("", GF_LOG_ERROR, "%s", msg); + gf_log (this->name, GF_LOG_ERROR, "%s", msg); *op_errstr = gf_strdup (msg); ret = -1; @@ -283,10 +288,10 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, ret = glusterd_check_gsync_running (volinfo, &is_run); if (ret && (is_run == _gf_false)) - gf_log ("", GF_LOG_WARNING, "Unable to get the status" + gf_log (this->name, GF_LOG_WARNING, "Unable to get the status" " of active "GEOREP" session"); if (is_run) { - gf_log ("", GF_LOG_WARNING, GEOREP" sessions active" + gf_log (this->name, GF_LOG_WARNING, GEOREP" sessions active" "for the volume %s ", volname); snprintf (msg, sizeof(msg), GEOREP" sessions are active " "for the volume %s.\nStop "GEOREP " sessions " @@ -301,29 +306,58 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, if (glusterd_is_defrag_on(volinfo)) { snprintf (msg, sizeof(msg), "Volume name %s rebalance is in " "progress. Please retry after completion", volname); - gf_log ("glusterd", GF_LOG_ERROR, "%s", msg); + gf_log (this->name, GF_LOG_ERROR, "%s", msg); *op_errstr = gf_strdup (msg); ret = -1; goto out; } + ctx = glusterd_op_get_ctx(); + switch (replace_op) { case GF_REPLACE_OP_START: if (glusterd_is_rb_started (volinfo)) { - gf_log ("", GF_LOG_ERROR, "Replace brick is already " - "started for volume "); + snprintf (msg, sizeof (msg), "Replace brick is already " + "started for volume"); + gf_log (this->name, GF_LOG_ERROR, msg); + *op_errstr = gf_strdup (msg); ret = -1; goto out; } + if (is_origin_glusterd ()) { + if (!ctx) { + ret = -1; + gf_log (this->name, GF_LOG_ERROR, + "Failed to get op_ctx"); + goto out; + } + + ret = glusterd_generate_and_set_task_id + (ctx, GF_REPLACE_BRICK_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_REPLACE_BRICK_TID_KEY, + &task_id_str); + if (ret) { + gf_log (this->name, GF_LOG_WARNING, + "Missing replace-brick-id"); + ret = 0; + } + } break; case GF_REPLACE_OP_PAUSE: if (glusterd_is_rb_paused (volinfo)) { - gf_log ("", GF_LOG_ERROR, "Replace brick is already " - "paused for volume "); + gf_log (this->name, GF_LOG_ERROR, "Replace brick is " + "already paused for volume "); ret = -1; goto out; } else if (!glusterd_is_rb_started(volinfo)) { - gf_log ("", GF_LOG_ERROR, "Replace brick is not" + gf_log (this->name, GF_LOG_ERROR, "Replace brick is not" " started for volume "); ret = -1; goto out; @@ -332,7 +366,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, case GF_REPLACE_OP_ABORT: if (!glusterd_is_rb_ongoing (volinfo)) { - gf_log ("", GF_LOG_ERROR, "Replace brick is not" + gf_log (this->name, GF_LOG_ERROR, "Replace brick is not" " started or paused for volume "); ret = -1; goto out; @@ -341,7 +375,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, case GF_REPLACE_OP_COMMIT: if (!glusterd_is_rb_ongoing (volinfo)) { - gf_log ("", GF_LOG_ERROR, "Replace brick is not " + gf_log (this->name, GF_LOG_ERROR, "Replace brick is not " "started for volume "); ret = -1; goto out; @@ -361,7 +395,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, goto out; } - gf_log (THIS->name, GF_LOG_ERROR, "%s", *op_errstr); + gf_log (this->name, GF_LOG_ERROR, "%s", *op_errstr); ret = -1; goto out; } @@ -381,10 +415,9 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, goto out; } - ctx = glusterd_op_get_ctx(); if (ctx) { if (!glusterd_is_fuse_available ()) { - gf_log ("glusterd", GF_LOG_ERROR, "Unable to open /dev/" + gf_log (this->name, GF_LOG_ERROR, "Unable to open /dev/" "fuse (%s), replace-brick command failed", strerror (errno)); snprintf (msg, sizeof(msg), "Fuse unavailable\n " @@ -396,7 +429,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, } if (!glusterd_is_local_addr (src_brickinfo->hostname)) { - gf_log ("", GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_DEBUG, "I AM THE SOURCE HOST"); if (src_brickinfo->port && rsp_dict) { ret = dict_set_int32 (rsp_dict, "src-brick-port", @@ -428,14 +461,14 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, dup_dstbrick = gf_strdup (dst_brick); if (!dup_dstbrick) { ret = -1; - gf_log ("", GF_LOG_ERROR, "Memory allocation failed"); + gf_log (this->name, GF_LOG_ERROR, "Memory allocation failed"); goto out; } host = strtok_r (dup_dstbrick, ":", &savetok); path = strtok_r (NULL, ":", &savetok); if (!host || !path) { - gf_log ("", GF_LOG_ERROR, + gf_log (this->name, GF_LOG_ERROR, "dst brick %s is not of form <HOSTNAME>:<export-dir>", dst_brick); ret = -1; @@ -451,7 +484,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, if (ret) { *op_errstr = gf_strdup (msg); ret = -1; - gf_log (THIS->name, GF_LOG_ERROR, *op_errstr); + gf_log (this->name, GF_LOG_ERROR, *op_errstr); goto out; } @@ -459,8 +492,8 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, (replace_op == GF_REPLACE_OP_START || replace_op == GF_REPLACE_OP_COMMIT_FORCE)) { - volinfo->src_brick = src_brickinfo; - volinfo->dst_brick = dst_brickinfo; + volinfo->rep_brick.src_brick = src_brickinfo; + volinfo->rep_brick.dst_brick = dst_brickinfo; } if (glusterd_rb_check_bricks (volinfo, src_brickinfo, dst_brickinfo)) { @@ -469,7 +502,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, *op_errstr = gf_strdup ("Incorrect source or " "destination brick"); if (*op_errstr) - gf_log (THIS->name, GF_LOG_ERROR, "%s", *op_errstr); + gf_log (this->name, GF_LOG_ERROR, "%s", *op_errstr); goto out; } @@ -511,7 +544,7 @@ glusterd_op_stage_replace_brick (dict_t *dict, char **op_errstr, out: GF_FREE (dup_dstbrick); - gf_log ("", GF_LOG_DEBUG, "Returning %d", ret); + gf_log (this->name, GF_LOG_DEBUG, "Returning %d", ret); return ret; } @@ -1499,6 +1532,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) char *dst_brick = NULL; glusterd_brickinfo_t *src_brickinfo = NULL; glusterd_brickinfo_t *dst_brickinfo = NULL; + char *task_id_str = NULL; this = THIS; GF_ASSERT (this); @@ -1508,26 +1542,23 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = dict_get_str (dict, "src-brick", &src_brick); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get src brick"); + gf_log (this->name, GF_LOG_ERROR, "Unable to get src brick"); goto out; } - gf_log (this->name, GF_LOG_DEBUG, - "src brick=%s", src_brick); + gf_log (this->name, GF_LOG_DEBUG, "src brick=%s", src_brick); ret = dict_get_str (dict, "dst-brick", &dst_brick); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get dst brick"); + gf_log (this->name, GF_LOG_ERROR, "Unable to get dst brick"); goto out; } - gf_log (this->name, GF_LOG_DEBUG, - "dst brick=%s", dst_brick); + gf_log (this->name, GF_LOG_DEBUG, "dst brick=%s", dst_brick); ret = dict_get_str (dict, "volname", &volname); - if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get volume name"); + gf_log (this->name, GF_LOG_ERROR, "Unable to get volume name"); goto out; } @@ -1540,28 +1571,30 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = glusterd_volinfo_find (volname, &volinfo); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to allocate memory"); + gf_log (this->name, GF_LOG_ERROR, "Unable to allocate memory"); goto out; } ret = glusterd_volume_brickinfo_get_by_brick (src_brick, volinfo, &src_brickinfo); if (ret) { - gf_log ("", GF_LOG_DEBUG, "Unable to get src-brickinfo"); + gf_log (this->name, GF_LOG_DEBUG, + "Unable to get src-brickinfo"); goto out; } ret = glusterd_get_rb_dst_brickinfo (volinfo, &dst_brickinfo); if (ret) { - gf_log ("", GF_LOG_ERROR, "Unable to get " + gf_log (this->name, GF_LOG_ERROR, "Unable to get " "replace brick destination brickinfo"); goto out; } ret = glusterd_resolve_brick (dst_brickinfo); if (ret) { - gf_log ("", GF_LOG_DEBUG, "Unable to resolve dst-brickinfo"); + gf_log (this->name, GF_LOG_DEBUG, + "Unable to resolve dst-brickinfo"); goto out; } @@ -1570,29 +1603,62 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) if (ret) goto out; + ctx = glusterd_op_get_ctx(); + if (!ctx) { + gf_log (this->name, GF_LOG_ERROR, "Failed to get op_ctx"); + ret = -1; + goto out; + } + if ((GF_REPLACE_OP_START != replace_op)) { ret = rb_update_dstbrick_port (dst_brickinfo, rsp_dict, dict, replace_op); if (ret) goto out; + + /* Set task-id, if available, in op_ctx dict for operations + * other than start + */ + if (!uuid_is_null (volinfo->rep_brick.rb_id)) { + ret = glusterd_copy_uuid_to_dict + (volinfo->rep_brick.rb_id, ctx, + GF_REPLACE_BRICK_TID_KEY); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "Failed to set replace-brick-id"); + goto out; + } + } } switch (replace_op) { case GF_REPLACE_OP_START: { + ret = dict_get_str (dict, GF_REPLACE_BRICK_TID_KEY, &task_id_str); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "Missing replace-brick-id"); + ret = 0; + } else { + uuid_parse (task_id_str, volinfo->rep_brick.rb_id); + } + if (!glusterd_is_local_addr (dst_brickinfo->hostname)) { - gf_log ("", GF_LOG_INFO, + gf_log (this->name, GF_LOG_INFO, "I AM THE DESTINATION HOST"); if (!glusterd_is_rb_paused (volinfo)) { - ret = rb_spawn_destination_brick (volinfo, dst_brickinfo); + ret = rb_spawn_destination_brick + (volinfo, dst_brickinfo); if (ret) { - gf_log ("", GF_LOG_DEBUG, - "Failed to spawn destination brick"); + gf_log (this->name, GF_LOG_DEBUG, + "Failed to spawn destination " + "brick"); goto out; } } else { - gf_log ("", GF_LOG_ERROR, "Replace brick is already " - "started=> no need to restart dst brick "); + gf_log (this->name, GF_LOG_ERROR, + "Replace brick is already started=> no " + "need to restart dst brick "); } } @@ -1601,14 +1667,14 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = rb_src_brick_restart (volinfo, src_brickinfo, 1); if (ret) { - gf_log ("", GF_LOG_DEBUG, - "Could not restart src-brick"); + gf_log (this->name, GF_LOG_DEBUG, + "Could not restart src-brick"); goto out; } } if (!glusterd_is_local_addr (dst_brickinfo->hostname)) { - gf_log ("", GF_LOG_INFO, + gf_log (this->name, GF_LOG_INFO, "adding dst-brick port no"); ret = rb_update_dstbrick_port (dst_brickinfo, rsp_dict, @@ -1629,7 +1695,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) dst_brickinfo, GF_REPLACE_OP_COMMIT); if (ret) { - gf_log ("", GF_LOG_ERROR, + gf_log (this->name, GF_LOG_ERROR, "Commit operation failed"); goto out; } @@ -1639,16 +1705,17 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) case GF_REPLACE_OP_COMMIT_FORCE: { ret = dict_set_int32 (volinfo->dict, "enable-pump", 0); - gf_log (THIS->name, GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_DEBUG, "Received commit - will be adding dst brick and " "removing src brick"); if (!glusterd_is_local_addr (dst_brickinfo->hostname)) { - gf_log (THIS->name, GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_DEBUG, "I AM THE DESTINATION HOST"); - ret = rb_kill_destination_brick (volinfo, dst_brickinfo); + ret = rb_kill_destination_brick (volinfo, + dst_brickinfo); if (ret) { - gf_log (THIS->name, GF_LOG_CRITICAL, + gf_log (this->name, GF_LOG_CRITICAL, "Unable to cleanup dst brick"); goto out; } @@ -1656,39 +1723,41 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = glusterd_nodesvcs_stop (volinfo); if (ret) { - gf_log (THIS->name, GF_LOG_ERROR, + gf_log (this->name, GF_LOG_ERROR, "Unable to stop nfs server, ret: %d", ret); } ret = glusterd_op_perform_replace_brick (volinfo, src_brick, dst_brick); if (ret) { - gf_log (THIS->name, GF_LOG_CRITICAL, "Unable to add " - "dst-brick: %s to volume: %s", - dst_brick, volinfo->volname); + gf_log (this->name, GF_LOG_CRITICAL, "Unable to add " + "dst-brick: %s to volume: %s", dst_brick, + volinfo->volname); (void) glusterd_nodesvcs_handle_graph_change (volinfo); goto out; } - volinfo->defrag_status = 0; + volinfo->rebal.defrag_status = 0; ret = glusterd_nodesvcs_handle_graph_change (volinfo); if (ret) { - gf_log (THIS->name, GF_LOG_CRITICAL, + gf_log (this->name, GF_LOG_CRITICAL, "Failed to generate nfs volume file"); } ret = glusterd_fetchspec_notify (THIS); glusterd_set_rb_status (volinfo, GF_RB_STATUS_NONE); - glusterd_brickinfo_delete (volinfo->dst_brick); - volinfo->src_brick = volinfo->dst_brick = NULL; + glusterd_brickinfo_delete (volinfo->rep_brick.dst_brick); + volinfo->rep_brick.src_brick = NULL; + volinfo->rep_brick.dst_brick = NULL; + uuid_clear (volinfo->rep_brick.rb_id); } break; case GF_REPLACE_OP_PAUSE: { - gf_log ("", GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_DEBUG, "Received pause - doing nothing"); ctx = glusterd_op_get_ctx (); if (ctx) { @@ -1696,7 +1765,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) dst_brickinfo, GF_REPLACE_OP_PAUSE); if (ret) { - gf_log ("", GF_LOG_ERROR, + gf_log (this->name, GF_LOG_ERROR, "Pause operation failed"); goto out; } @@ -1715,7 +1784,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) dst_brickinfo, GF_REPLACE_OP_ABORT); if (ret) { - gf_log (THIS->name, GF_LOG_ERROR, + gf_log (this->name, GF_LOG_ERROR, "Abort operation failed"); goto out; } @@ -1723,7 +1792,8 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = dict_set_int32 (volinfo->dict, "enable-pump", 0); if (ret) { - gf_log (THIS->name, GF_LOG_CRITICAL, "Unable to disable pump"); + gf_log (this->name, GF_LOG_CRITICAL, + "Unable to disable pump"); } if (!glusterd_is_local_addr (src_brickinfo->hostname)) { @@ -1738,7 +1808,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) } if (!glusterd_is_local_addr (dst_brickinfo->hostname)) { - gf_log (THIS->name, GF_LOG_INFO, + gf_log (this->name, GF_LOG_INFO, "I AM THE DESTINATION HOST"); ret = rb_kill_destination_brick (volinfo, dst_brickinfo); if (ret) { @@ -1748,14 +1818,15 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) } } glusterd_set_rb_status (volinfo, GF_RB_STATUS_NONE); - glusterd_brickinfo_delete (volinfo->dst_brick); - volinfo->src_brick = volinfo->dst_brick = NULL; + glusterd_brickinfo_delete (volinfo->rep_brick.dst_brick); + volinfo->rep_brick.src_brick = NULL; + volinfo->rep_brick.dst_brick = NULL; } break; case GF_REPLACE_OP_STATUS: { - gf_log ("", GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_DEBUG, "received status - doing nothing"); ctx = glusterd_op_get_ctx (); if (ctx) { @@ -1763,7 +1834,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = dict_set_str (ctx, "status-reply", "replace brick has been paused"); if (ret) - gf_log (THIS->name, GF_LOG_ERROR, + gf_log (this->name, GF_LOG_ERROR, "failed to set pump status" " in ctx"); goto out; @@ -1787,7 +1858,7 @@ glusterd_op_replace_brick (dict_t *dict, dict_t *rsp_dict) ret = glusterd_store_volinfo (volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT); if (ret) - gf_log (THIS->name, GF_LOG_ERROR, "Couldn't store" + gf_log (this->name, GF_LOG_ERROR, "Couldn't store" " replace brick operation's state"); out: |