[checkmk-commits] Check_MK Git: check_mk: netapp_api_volumes:<br>

Module: check_mk Branch: master Commit: db8d3a12d6ea80183c06f7ac0f4ead2da19da44e URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=db8d3a12d6ea80… Author: Andreas Boesl &lt;ab(a)mathias-kettner.de&gt; Date: Tue Apr 28 16:55:17 2015 +0200 netapp_api_volumes:<br> fixed misleading warning for offline volumes, now also reports inode information netapp_api_disks:<br> reformatted check output, fixed incorrect disk numbers, data and parity disk now have a separate prefailed information. Conflicts: ChangeLog checks/netapp_api_volumes --- .werks/2072 | 16 ++++++++++++++ ChangeLog | 1 + agents/special/agent_netapp | 10 +++++---- checks/netapp_api_disk | 50 ++++++++++++++++++++++++++----------------- checks/netapp_api_volumes | 10 +++++++-- 5 files changed, 61 insertions(+), 26 deletions(-) diff --git a/.werks/2072 b/.werks/2072 new file mode 100644 index 0000000..85eabf4 --- /dev/null +++ b/.werks/2072 @@ -0,0 +1,16 @@ +Title: agent_netapp, netapp_api_volumes, netapp_api_disks: Improved check output +Level: 1 +Component: checks +Class: fix +Compatible: compat +State: unknown +Version: 1.2.7i1 +Date: 1430232561 + +netapp_api_volumes:<br> +fixed misleading warning for offline volumes, now also reports inode information + + +netapp_api_disks:<br> +reformatted check output, fixed incorrect disk numbers, data and parity disk now have a separate prefailed information. + diff --git a/ChangeLog b/ChangeLog index 8a0ed7f..4a2b5ba 100644 --- a/ChangeLog +++ b/ChangeLog @@ -348,6 +348,7 @@ * 2234 FIX: AIX Agent: Forced load of environment and fixed call of trd (Libelle Business Shadow)... * 2247 FIX: ibm_svc_mdiskgrp: Made check working with different firmware versions outputs * 2071 FIX: windows agent: fileinfo check now correctly reports empty directories... + * 2072 FIX: agent_netapp, netapp_api_volumes, netapp_api_disks: Improved check output... Multisite: * 1758 Improved exception hander: Shows details without additional debug request, added mailto link for error report... diff --git a/agents/special/agent_netapp b/agents/special/agent_netapp index 83c8947..32147d1 100755 --- a/agents/special/agent_netapp +++ b/agents/special/agent_netapp @@ -55,7 +55,8 @@ netapp_objects["netapp_api_cpu"] = { } netapp_objects["netapp_api_volumes"] = { - "configs" : { "volume-list-info" : ["name", "volume-info", "size-total", "size-available", "volumes"] }, + "configs" : { "volume-list-info" : ["name", "volume-info", "size-total", "size-available", + "volumes", "files-total", "files-used", "state"] }, # Feel free to execute this construct in a python interpreter. It just creates a list of elements... "counters" : { "volume": sum(map(lambda x: ["%s" % x, "nfs_%s" % x, "cifs_%s" % x, "san_%s" % x, "fcp_%s" % x, "iscsi_%s" % x], sum(map(lambda x: ["read_%s" % x, "write_%s" % x], ["data", "latency"]), [])), []) + [ "instance_name" ] } @@ -72,6 +73,7 @@ netapp_objects["netapp_api_protocol"] = { "iscsi": ["instance_name", "iscsi_read_ops", "iscsi_write_ops"], "cifs": ["instance_name", "cifs_read_ops", "cifs_write_ops"], "fcp": ["instance_name", "fcp_read_ops", "fcp_write_data"], +# "system": [], } } @@ -89,6 +91,7 @@ netapp_objects["netapp_api_cluster"] = { } + def output_disks(results): bay_list = results["configs"]["storage-shelf-bay-list-info"].child_get("shelf-bay-list") shelf_uids = {} @@ -117,7 +120,8 @@ def output_disks(results): print "disk-detail-info" print "%s\t%s" % ("disk-uid", disk_uid) print "%s\t%s" % ("raid-state", raid_state) - for key in [ "bay", "used-space", "physical-space", "shelf", "serial-number" ]: + for key in [ "bay", "used-space", "physical-space", "shelf", "serial-number", + "is-prefailed", "raid-type" ]: print "%s\t%s" % (key, disk.child_get_string(key)) print "[config_instance]\tshelf-uids-of-disks" @@ -414,8 +418,6 @@ try: raise Exception("\n".join(errors)) except Exception, e: - if opt_debug: - raise sys.stderr.write("Error(s) on processing the received data:\n%s" % e) sys.exit(1) diff --git a/checks/netapp_api_disk b/checks/netapp_api_disk index 66ac334..4582ff6 100644 --- a/checks/netapp_api_disk +++ b/checks/netapp_api_disk @@ -60,33 +60,44 @@ def check_netapp_api_disk_summary(_no_item, params, info): my_disks = dict([disk for disk in disks.items() if not disk[1].get("raid-state") == "partner"]) spare_count = 0 - prefailed_disks = [] - broken_disks = [] + data_disks = [] + parity_disks = [] + broken_disks = [] + prefailed_data_disks = [] + prefailed_parity_disks = [] raid_states = {} - phys_space = 0 - total_space = 0 + phys_space = 0 + total_space = 0 for name, disk in my_disks.items(): - if disk.get("raid-state") == "partner": - continue # Disk belongs to partner - total_space += int(disk.get("used-space")) + raid_type = disk.get("raid-type") if disk.get("raid-state", "") == "broken": broken_disks.append(disk) - elif disk.get("is-prefailed", "false") != "false": - prefailed_disks.append(disk) + elif disk.get("is-prefailed", "false") not in [ "false", "None" ]: + if raid_type in ["parity", "dparity"]: + prefailed_parity_disks.append(disk) + elif raid_type == "data": + prefailed_data_disks.append(disk) elif disk.get("raid-state","") == "spare": spare_count += 1 - if disk.get("raid-type"): - raid_states.setdefault(disk["raid-type"], 0) - raid_states[disk["raid-type"]] += 1 + + if raid_type in ["parity", "dparity"]: + parity_disks.append(disk) + elif raid_type in ["data"]: + data_disks.append(disk) yield 0, "Total Raw Capacity: %s" % get_bytes_human_readable(total_space), [("total_space", total_space)] - yield 0, "Disks Total: %s Spare: %s" % (len(my_disks), spare_count), [ ("total", len(my_disks)), - ("spare", spare_count), - ("broken", len(broken_disks)) ] + yield 0, "Total disks: %d (%d Spare)" % (len(my_disks) - len(broken_disks), spare_count), [ + ("total", len(my_disks)), + ("spare", spare_count), + ("broken", len(broken_disks)) ] + + yield 0, "Data disks: %d (%d prefailed)" % (len(data_disks), len(prefailed_data_disks)) + yield 0, "Parity disks: %d (%d prefailed)" % (len(parity_disks), len(prefailed_parity_disks)) + yield 0, "Broken disks: %d" % len(broken_disks) disk_to_shelf_map = {} @@ -98,8 +109,9 @@ def check_netapp_api_disk_summary(_no_item, params, info): disk_to_shelf_map[disk] = shelf return disk_to_shelf_map.get(disk_uid) - for text, disks, state in [("Prefailed", prefailed_disks, 0), - ("Broken" , broken_disks, 0)]: + for text, disks, state in [("Data-Prefailed", prefailed_data_disks, 0), + ("Parity-Prefailed", prefailed_parity_disks, 0), + ("Broken", broken_disks, 0)]: info = [] for disk in disks: disk_info = "Serial: %s" % disk.get("serial-number") @@ -111,9 +123,7 @@ def check_netapp_api_disk_summary(_no_item, params, info): info.append(disk_info) if info: - yield state, "%s: %s - %s" % (text, len(disks), " / ".join(info)) - else: - yield 0, "%s: %s" % (text, len(disks)) + yield state, "%s Disk Details: %s" % (text, " / ".join(info)) if broken_disks: warn, crit = params.get("broken_spare_ratio", (1.0, 50.0)) diff --git a/checks/netapp_api_volumes b/checks/netapp_api_volumes index d6bb34d..62616ee 100644 --- a/checks/netapp_api_volumes +++ b/checks/netapp_api_volumes @@ -62,12 +62,18 @@ def check_netapp_api_volumes(item, params, parsed): volume = find_volume(item) if not volume: - return (3, "Volume not found in agent output") + return 3, "Volume not found in agent output" + + if volume.get("state") == "offline": + return 1, "Volume is offline" mega = 1024.0 * 1024.0 size_total = int(volume.get("size-total")) / mega size_avail = int(volume.get("size-available")) / mega - state, info, perf = df_check_filesystem_list(item, params, [(volume.get("name"), size_total, size_avail, 0)]) + inodes_total = int(volume.get("files-total")) + inodes_avail = inodes_total - int(volume.get("files-used")) + state, info, perf = df_check_filesystem_single(g_hostname, item, size_total, size_avail, 0, + inodes_total, inodes_avail, params) counter_wrapped = False counters = []