diff options
Diffstat (limited to 'extras/snap_scheduler/snap_scheduler.py')
| -rwxr-xr-x | extras/snap_scheduler/snap_scheduler.py | 307 |
1 files changed, 262 insertions, 45 deletions
diff --git a/extras/snap_scheduler/snap_scheduler.py b/extras/snap_scheduler/snap_scheduler.py index af092e2c341..e8fcc449a9b 100755 --- a/extras/snap_scheduler/snap_scheduler.py +++ b/extras/snap_scheduler/snap_scheduler.py @@ -1,4 +1,4 @@ -#!/usr/bin/env python +#!/usr/bin/python3 # # Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> # This file is part of GlusterFS. @@ -19,12 +19,55 @@ import logging.handlers import sys import shutil from errno import EEXIST - +from conf import GLUSTERFS_LIBEXECDIR +sys.path.insert(1, GLUSTERFS_LIBEXECDIR) + +EVENTS_ENABLED = True +try: + from events.eventtypes import SNAPSHOT_SCHEDULER_INITIALISED \ + as EVENT_SNAPSHOT_SCHEDULER_INITIALISED + from events.eventtypes import SNAPSHOT_SCHEDULER_INIT_FAILED \ + as EVENT_SNAPSHOT_SCHEDULER_INIT_FAILED + from events.eventtypes import SNAPSHOT_SCHEDULER_DISABLED \ + as EVENT_SNAPSHOT_SCHEDULER_DISABLED + from events.eventtypes import SNAPSHOT_SCHEDULER_DISABLE_FAILED \ + as EVENT_SNAPSHOT_SCHEDULER_DISABLE_FAILED + from events.eventtypes import SNAPSHOT_SCHEDULER_ENABLED \ + as EVENT_SNAPSHOT_SCHEDULER_ENABLED + from events.eventtypes import SNAPSHOT_SCHEDULER_ENABLE_FAILED \ + as EVENT_SNAPSHOT_SCHEDULER_ENABLE_FAILED + from events.eventtypes import SNAPSHOT_SCHEDULER_SCHEDULE_ADDED \ + as EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_ADDED + from events.eventtypes import SNAPSHOT_SCHEDULER_SCHEDULE_ADD_FAILED \ + as EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_ADD_FAILED + from events.eventtypes import SNAPSHOT_SCHEDULER_SCHEDULE_DELETED \ + as EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_DELETED + from events.eventtypes import SNAPSHOT_SCHEDULER_SCHEDULE_DELETE_FAILED \ + as EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_DELETE_FAILED + from events.eventtypes import SNAPSHOT_SCHEDULER_SCHEDULE_EDITED \ + as EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_EDITED + from events.eventtypes import SNAPSHOT_SCHEDULER_SCHEDULE_EDIT_FAILED \ + as EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_EDIT_FAILED +except ImportError: + # Events APIs not installed, dummy eventtypes with None + EVENTS_ENABLED = False + EVENT_SNAPSHOT_SCHEDULER_INITIALISED = None + EVENT_SNAPSHOT_SCHEDULER_INIT_FAILED = None + EVENT_SNAPSHOT_SCHEDULER_DISABLED = None + EVENT_SNAPSHOT_SCHEDULER_DISABLE_FAILED = None + EVENT_SNAPSHOT_SCHEDULER_ENABLED = None + EVENT_SNAPSHOT_SCHEDULER_ENABLE_FAILED = None + EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_ADDED = None + EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_ADD_FAILED = None + EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_DELETED = None + EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_DELETE_FAILED = None + EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_EDITED = None + EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_EDIT_FAILED = None SCRIPT_NAME = "snap_scheduler" scheduler_enabled = False log = logging.getLogger(SCRIPT_NAME) -SHARED_STORAGE_DIR="/var/run/gluster/shared_storage" +SHARED_STORAGE_DIR="/run/gluster/shared_storage" GCRON_DISABLED = SHARED_STORAGE_DIR+"/snaps/gcron_disabled" GCRON_ENABLED = SHARED_STORAGE_DIR+"/snaps/gcron_enabled" GCRON_TASKS = SHARED_STORAGE_DIR+"/snaps/glusterfs_snap_cron_tasks" @@ -55,6 +98,42 @@ INVALID_SCHEDULE = 15 INVALID_ARG = 16 VOLUME_DOES_NOT_EXIST = 17 +def print_error (error_num): + if error_num == INTERNAL_ERROR: + return "Internal Error" + elif error_num == SHARED_STORAGE_DIR_DOESNT_EXIST: + return "The shared storage directory ("+SHARED_STORAGE_DIR+")" \ + " does not exist." + elif error_num == SHARED_STORAGE_NOT_MOUNTED: + return "The shared storage directory ("+SHARED_STORAGE_DIR+")" \ + " is not mounted." + elif error_num == ANOTHER_TRANSACTION_IN_PROGRESS: + return "Another transaction is in progress." + elif error_num == INIT_FAILED: + return "Initialisation failed." + elif error_num == SCHEDULING_ALREADY_DISABLED: + return "Snapshot scheduler is already disabled." + elif error_num == SCHEDULING_ALREADY_ENABLED: + return "Snapshot scheduler is already enabled." + elif error_num == NODE_NOT_INITIALISED: + return "The node is not initialised." + elif error_num == ANOTHER_SCHEDULER_ACTIVE: + return "Another scheduler is active." + elif error_num == JOB_ALREADY_EXISTS: + return "The job already exists." + elif error_num == JOB_NOT_FOUND: + return "The job cannot be found." + elif error_num == INVALID_JOBNAME: + return "The job name is invalid." + elif error_num == INVALID_VOLNAME: + return "The volume name is invalid." + elif error_num == INVALID_SCHEDULE: + return "The schedule is invalid." + elif error_num == INVALID_ARG: + return "The argument is invalid." + elif error_num == VOLUME_DOES_NOT_EXIST: + return "The volume does not exist." + def output(msg): print("%s: %s" % (SCRIPT_NAME, msg)) @@ -70,7 +149,7 @@ def initLogger(): sh.setFormatter(formatter) process = subprocess.Popen(["gluster", "--print-logdir"], - stdout=subprocess.PIPE) + stdout=subprocess.PIPE, universal_newlines=True) logfile = os.path.join(process.stdout.read()[:-1], SCRIPT_NAME + ".log") fh = logging.FileHandler(logfile) @@ -128,11 +207,11 @@ def enable_scheduler(): os.remove(GCRON_TASKS) try: f = os.open(GCRON_ENABLED, os.O_CREAT | os.O_NONBLOCK, - 0644) + 0o644) os.close(f) - except OSError as (errno, strerror): + except OSError as e: log.error("Failed to open %s. Error: %s.", - GCRON_ENABLED, strerror) + GCRON_ENABLED, e) ret = INTERNAL_ERROR return ret os.symlink(GCRON_ENABLED, GCRON_TASKS) @@ -140,8 +219,9 @@ def enable_scheduler(): log.info("Snapshot scheduling is enabled") output("Snapshot scheduling is enabled") ret = 0 - except OSError as (errno, strerror): - print_str = "Failed to enable snapshot scheduling. Error: "+strerror + except OSError as e: + print_str = ("Failed to enable snapshot scheduling." + "Error: {{}}" + e) log.error(print_str) output(print_str) ret = INTERNAL_ERROR @@ -183,14 +263,15 @@ def disable_scheduler(): os.remove(GCRON_DISABLED) if os.path.lexists(GCRON_TASKS): os.remove(GCRON_TASKS) - f = os.open(GCRON_DISABLED, os.O_CREAT, 0644) + f = os.open(GCRON_DISABLED, os.O_CREAT, 0o644) os.close(f) os.symlink(GCRON_DISABLED, GCRON_TASKS) log.info("Snapshot scheduling is disabled") output("Snapshot scheduling is disabled") ret = 0 - except OSError as (errno, strerror): - print_str = "Failed to disable snapshot scheduling. Error: "+strerror + except OSError as e: + print_str = ("Failed to disable snapshot scheduling. Error: " + + e) log.error(print_str) output(print_str) ret = INTERNAL_ERROR @@ -229,8 +310,8 @@ def load_tasks_from_file(): tasks[jobname] = schedule+":"+volname f.close() ret = 0 - except IOError as (errno, strerror): - log.error("Failed to open %s. Error: %s.", GCRON_ENABLED, strerror) + except IOError as e: + log.error("Failed to open %s. Error: %s.", GCRON_ENABLED, e) ret = INTERNAL_ERROR return ret @@ -243,8 +324,8 @@ def get_current_scheduler(): current_scheduler = f.readline().rstrip('\n') f.close() ret = 0 - except IOError as (errno, strerror): - log.error("Failed to open %s. Error: %s.", CURRENT_SCHEDULER, strerror) + except IOError as e: + log.error("Failed to open %s. Error: %s.", CURRENT_SCHEDULER, e) ret = INTERNAL_ERROR return ret @@ -284,7 +365,7 @@ def list_schedules(): def write_tasks_to_file(): try: - with open(TMP_FILE, "w", 0644) as f: + with open(TMP_FILE, "w", 0o644) as f: # If tasks is empty, just create an empty tmp file if len(tasks) != 0: for key in sorted(tasks): @@ -297,8 +378,8 @@ def write_tasks_to_file(): f.flush() os.fsync(f.fileno()) f.close() - except IOError as (errno, strerror): - log.error("Failed to open %s. Error: %s.", TMP_FILE, strerror) + except IOError as e: + log.error("Failed to open %s. Error: %s.", TMP_FILE, e) ret = INTERNAL_ERROR return ret @@ -309,13 +390,13 @@ def write_tasks_to_file(): def update_current_scheduler(data): try: - with open(TMP_FILE, "w", 0644) as f: + with open(TMP_FILE, "w", 0o644) as f: f.write("%s" % data) f.flush() os.fsync(f.fileno()) f.close() - except IOError as (errno, strerror): - log.error("Failed to open %s. Error: %s.", TMP_FILE, strerror) + except IOError as e: + log.error("Failed to open %s. Error: %s.", TMP_FILE, e) ret = INTERNAL_ERROR return ret @@ -378,11 +459,11 @@ def add_schedules(jobname, schedule, volname): job_lockfile = LOCK_FILE_DIR + jobname try: f = os.open(job_lockfile, os.O_CREAT | os.O_NONBLOCK, - 0644) + 0o644) os.close(f) - except OSError as (errno, strerror): + except OSError as e: log.error("Failed to open %s. Error: %s.", - job_lockfile, strerror) + job_lockfile, e) ret = INTERNAL_ERROR return ret log.info("Successfully added snapshot schedule %s" % @@ -410,9 +491,9 @@ def delete_schedules(jobname): job_lockfile = LOCK_FILE_DIR+jobname try: os.remove(job_lockfile) - except OSError as (errno, strerror): + except OSError as e: log.error("Failed to open %s. Error: %s.", - job_lockfile, strerror) + job_lockfile, e) ret = INTERNAL_ERROR return ret log.info("Successfully deleted snapshot schedule %s" @@ -466,18 +547,113 @@ def edit_schedules(jobname, schedule, volname): return ret +def get_bool_val(): + getsebool_cli = ["getsebool", + "-a"] + p1 = subprocess.Popen(getsebool_cli, stdout=subprocess.PIPE, + stderr=subprocess.PIPE) + + grep_cmd = ["grep", + "cron_system_cronjob_use_shares"] + p2 = subprocess.Popen(grep_cmd, stdin=p1.stdout, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE) + + p1.stdout.close() + output, err = p2.communicate() + rv = p2.returncode + + if rv: + log.error("Command output:") + log.error(err) + return -1 + + bool_val = output.split()[2] + log.debug("Bool value = '%s'", bool_val) + + return bool_val + +def get_selinux_status(): + getenforce_cli = ["getenforce"] + log.debug("Running command '%s'", " ".join(getenforce_cli)) + + try: + p1 = subprocess.Popen(getenforce_cli, stdout=subprocess.PIPE, + stderr=subprocess.PIPE) + except OSError as oserr: + log.error("Failed to run the command \"getenforce\". Error: %s" %\ + oserr) + return -1 + + output, err = p1.communicate() + rv = p1.returncode + + if rv: + log.error("Command output:") + log.error(err) + return -1 + else: + selinux_status=output.rstrip() + log.debug("selinux status: %s", selinux_status) + + return selinux_status + +def set_cronjob_user_share(): + selinux_status = get_selinux_status() + if (selinux_status == -1): + log.error("Failed to get selinux status") + return -1 + elif (selinux_status == "Disabled"): + return 0 + + bool_val = get_bool_val() + # In case of a failure (where the boolean value is not) + # present in the system, we should not proceed further + # We should only proceed when the value is "off" + if (bool_val == -1 or bool_val != "off"): + return 0 + + setsebool_cli = ["setsebool", "-P", + "cron_system_cronjob_use_shares", + "on"] + log.debug("Running command '%s'", " ".join(setsebool_cli)) + + p1 = subprocess.Popen(setsebool_cli, stdout=subprocess.PIPE, + stderr=subprocess.PIPE) + + output, err = p1.communicate() + rv = p1.returncode + + if rv: + log.error("Command output:") + log.error(err) + return rv + + bool_val = get_bool_val() + if (bool_val == "on"): + return 0 + else: + # In case of an error or if boolean is not on + # we return a failure here + return -1 def initialise_scheduler(): + ret = set_cronjob_user_share() + if ret: + log.error("Failed to set selinux boolean " + "cron_system_cronjob_use_shares to 'on'") + return ret + try: - with open(TMP_FILE, "w+", 0644) as f: + with open(TMP_FILE, "w+", 0o644) as f: updater = ("* * * * * root PATH=$PATH:/usr/local/sbin:" "/usr/sbin gcron.py --update\n") f.write("%s\n" % updater) f.flush() os.fsync(f.fileno()) f.close() - except IOError as (errno, strerror): - log.error("Failed to open %s. Error: %s.", TMP_FILE, strerror) + except IOError as e: + log.error("Failed to open %s. Error: %s.", TMP_FILE, e) ret = INIT_FAILED return ret @@ -485,10 +661,10 @@ def initialise_scheduler(): if not os.path.lexists(GCRON_TASKS): try: - f = open(GCRON_TASKS, "w", 0644) + f = open(GCRON_TASKS, "w", 0o644) f.close() - except IOError as (errno, strerror): - log.error("Failed to open %s. Error: %s.", GCRON_TASKS, strerror) + except IOError as e: + log.error("Failed to open %s. Error: %s.", GCRON_TASKS, e) ret = INIT_FAILED return ret @@ -499,6 +675,7 @@ def initialise_scheduler(): log.info("Successfully initialised snapshot scheduler for this node") output("Successfully initialised snapshot scheduler for this node") + gf_event (EVENT_SNAPSHOT_SCHEDULER_INITIALISED, status="Success") ret = 0 return ret @@ -545,6 +722,8 @@ def perform_operation(args): ret = initialise_scheduler() if ret != 0: output("Failed to initialise snapshot scheduling") + gf_event (EVENT_SNAPSHOT_SCHEDULER_INIT_FAILED, + error=print_error(ret)) return ret # Disable snapshot scheduler @@ -552,6 +731,11 @@ def perform_operation(args): ret = disable_scheduler() if ret == 0: subprocess.Popen(["touch", "-h", GCRON_TASKS]) + gf_event (EVENT_SNAPSHOT_SCHEDULER_DISABLED, + status="Successfully Disabled") + else: + gf_event (EVENT_SNAPSHOT_SCHEDULER_DISABLE_FAILED, + error=print_error(ret)) return ret # Check if the symlink to GCRON_TASKS is properly set in the shared storage @@ -582,6 +766,11 @@ def perform_operation(args): ret = enable_scheduler() if ret == 0: subprocess.Popen(["touch", "-h", GCRON_TASKS]) + gf_event (EVENT_SNAPSHOT_SCHEDULER_ENABLED, + status="Successfully Enabled") + else: + gf_event (EVENT_SNAPSHOT_SCHEDULER_ENABLE_FAILED, + error=print_error(ret)) return ret # Disable snapshot scheduler @@ -589,6 +778,11 @@ def perform_operation(args): ret = disable_scheduler() if ret == 0: subprocess.Popen(["touch", "-h", GCRON_TASKS]) + gf_event (EVENT_SNAPSHOT_SCHEDULER_DISABLED, + status="Successfully Disabled") + else: + gf_event (EVENT_SNAPSHOT_SCHEDULER_DISABLE_FAILED, + error=print_error(ret)) return ret # List snapshot schedules @@ -604,6 +798,12 @@ def perform_operation(args): ret = add_schedules(args.jobname, args.schedule, args.volname) if ret == 0: subprocess.Popen(["touch", "-h", GCRON_TASKS]) + gf_event (EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_ADDED, + status="Successfully added job "+args.jobname) + else: + gf_event (EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_ADD_FAILED, + status="Failed to add job "+args.jobname, + error=print_error(ret)) return ret # Delete snapshot schedules @@ -614,6 +814,12 @@ def perform_operation(args): ret = delete_schedules(args.jobname) if ret == 0: subprocess.Popen(["touch", "-h", GCRON_TASKS]) + gf_event (EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_DELETED, + status="Successfully deleted job "+args.jobname) + else: + gf_event (EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_DELETE_FAILED, + status="Failed to delete job "+args.jobname, + error=print_error(ret)) return ret # Edit snapshot schedules @@ -624,11 +830,22 @@ def perform_operation(args): ret = edit_schedules(args.jobname, args.schedule, args.volname) if ret == 0: subprocess.Popen(["touch", "-h", GCRON_TASKS]) + gf_event (EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_EDITED, + status="Successfully edited job "+args.jobname) + else: + gf_event (EVENT_SNAPSHOT_SCHEDULER_SCHEDULE_EDIT_FAILED, + status="Failed to edit job "+args.jobname, + error=print_error(ret)) return ret ret = INVALID_ARG return ret +def gf_event(event_type, **kwargs): + if EVENTS_ENABLED: + from events.gf_event import gf_event as gfevent + gfevent(event_type, **kwargs) + def main(argv): initLogger() @@ -679,42 +896,42 @@ def main(argv): if not os.path.exists(SHARED_STORAGE_DIR+"/snaps/"): try: os.makedirs(SHARED_STORAGE_DIR+"/snaps/") - except OSError as (errno, strerror): + except OSError as e: if errno != EEXIST: - log.error("Failed to create %s : %s", SHARED_STORAGE_DIR+"/snaps/", strerror) + log.error("Failed to create %s : %s", SHARED_STORAGE_DIR+"/snaps/", e) output("Failed to create %s. Error: %s" - % (SHARED_STORAGE_DIR+"/snaps/", strerror)) + % (SHARED_STORAGE_DIR+"/snaps/", e)) return INTERNAL_ERROR if not os.path.exists(GCRON_ENABLED): - f = os.open(GCRON_ENABLED, os.O_CREAT | os.O_NONBLOCK, 0644) + f = os.open(GCRON_ENABLED, os.O_CREAT | os.O_NONBLOCK, 0o644) os.close(f) if not os.path.exists(LOCK_FILE_DIR): try: os.makedirs(LOCK_FILE_DIR) - except OSError as (errno, strerror): + except OSError as e: if errno != EEXIST: - log.error("Failed to create %s : %s", LOCK_FILE_DIR, strerror) + log.error("Failed to create %s : %s", LOCK_FILE_DIR, e) output("Failed to create %s. Error: %s" - % (LOCK_FILE_DIR, strerror)) + % (LOCK_FILE_DIR, e)) return INTERNAL_ERROR try: - f = os.open(LOCK_FILE, os.O_CREAT | os.O_RDWR | os.O_NONBLOCK, 0644) + f = os.open(LOCK_FILE, os.O_CREAT | os.O_RDWR | os.O_NONBLOCK, 0o644) try: fcntl.flock(f, fcntl.LOCK_EX | fcntl.LOCK_NB) ret = perform_operation(args) fcntl.flock(f, fcntl.LOCK_UN) - except IOError as (errno, strerror): + except IOError: log.info("%s is being processed by another agent.", LOCK_FILE) output("Another snap_scheduler command is running. " "Please try again after some time.") return ANOTHER_TRANSACTION_IN_PROGRESS os.close(f) - except OSError as (errno, strerror): - log.error("Failed to open %s : %s", LOCK_FILE, strerror) - output("Failed to open %s. Error: %s" % (LOCK_FILE, strerror)) + except OSError as e: + log.error("Failed to open %s : %s", LOCK_FILE, e) + output("Failed to open %s. Error: %s" % (LOCK_FILE, e)) return INTERNAL_ERROR return ret |