zulip/scripts/restart-server

#!/usr/bin/env python3
import logging
import os
import pwd
import shlex
import subprocess
import sys
import time

sys.path.append(os.path.join(os.path.dirname(__file__), ".."))
from scripts.lib.supervisor import list_supervisor_processes
from scripts.lib.zulip_tools import (
    DEPLOYMENTS_DIR,
    ENDC,
    OKGREEN,
    WARNING,
    get_config,
    get_config_file,
    get_tornado_ports,
    has_application_server,
    has_process_fts_updates,
    overwrite_symlink,
    start_arg_parser,
)

action = "restart"
if not sys.argv[0].endswith("restart-server"):
    action = "start"
verbing = action.title() + "ing"

logging.Formatter.converter = time.gmtime
logging.basicConfig(format=f"%(asctime)s {action}-server: %(message)s", level=logging.INFO)

parser = start_arg_parser(action=action, add_help=True)
args = parser.parse_args()

deploy_path = os.path.realpath(os.path.join(os.path.dirname(__file__), ".."))
os.chdir(deploy_path)

if pwd.getpwuid(os.getuid()).pw_name != "zulip":
    logging.error("Must be run as user 'zulip'.")
    sys.exit(1)

if args.fill_cache:
    logging.info("Filling memcached caches")
    subprocess.check_call(["./manage.py", "fill_memcached_caches"])

current_symlink = os.path.join(DEPLOYMENTS_DIR, "current")
last_symlink = os.path.join(DEPLOYMENTS_DIR, "last")
change_symlink = os.readlink(current_symlink) != deploy_path
if change_symlink:
    overwrite_symlink(os.readlink(current_symlink), last_symlink)
    overwrite_symlink(deploy_path, current_symlink)

config_file = get_config_file()
tornado_ports = get_tornado_ports(config_file)
workers = []

if has_application_server():
    # Start by restarting the workers and similar processes, one at a
    # time.  Workers can always support processing events with old event
    # contents, but cannot necessarily understand events enqueued by a
    # newer Django process.  Restarting them one at a time, rather than
    # all-at-once, minimizes the downtime of each, and reduces startup
    # contention.
    #
    # For "start" or less-graceful circumstances, we don't need to
    # iterate; we'll stop all of them at once, and start them all later.
    # In those cases, using the glob form is faster -- but if we do need
    # to iterate, we need to expand the glob.
    if action == "start" or args.less_graceful:
        workers.append("zulip-workers:*")
    else:
        workers.extend(list_supervisor_processes(["zulip-workers:*"]))

    if has_application_server(once=True):
        # These used to be included in "zulip-workers:*"; since we may
        # be restarting an older version of Zulip, which has not
        # applied puppet to reload the new list of processes, only
        # stop them if they currently exist according to
        # `supervisorctl`.
        workers.extend(
            list_supervisor_processes(
                [
                    "zulip_deliver_scheduled_emails",
                    "zulip_deliver_scheduled_messages",
                ]
            )
        )

if has_process_fts_updates():
    workers.append("process-fts-updates")

if action == "restart" and len(workers) > 0:
    if args.less_graceful:
        # The less graceful form stops every worker now; we start them
        # back up at the end.
        logging.info("Stopping workers")
        subprocess.check_call(["supervisorctl", "stop", *workers])
    else:
        # We cannot pass all of these to one `supervisorctl restart`
        # because that takes them all down at once, waits until they are
        # all down, and then brings them back up; doing them sequentially
        # requires multiple `supervisorctl restart` calls.
        for worker in workers:
            logging.info("Restarting %s", worker)
            subprocess.check_call(["supervisorctl", "restart", worker])

if has_application_server():
    # Next, we restart the Tornado processes sequentially, in order to
    # minimize downtime of the tornado service caused by too many Python
    # processes restarting at the same time, resulting in each receiving
    # insufficient priority.  This is important, because Tornado is the
    # main source of user-visible downtime when we restart a Zulip server.
    # We do this before restarting Django, in case there are new event
    # types which it will need to know how to deal with.
    if action == "restart" and args.skip_tornado:
        logging.info("Skipping restart of Tornado")
    elif len(tornado_ports) > 1:
        for p in tornado_ports:
            # Restart Tornado processes individually for a better rate of
            # restarts.  This also avoids behavior with restarting a whole
            # supervisord group where if any individual process is slow to
            # stop, the whole bundle stays stopped for an extended time.
            logging.info("%s Tornado process on port %s", verbing, p)
            subprocess.check_call(
                ["supervisorctl", action, f"zulip-tornado:zulip-tornado-port-{p}"]
            )
    else:
        logging.info("%s Tornado process", verbing)
        subprocess.check_call(["supervisorctl", action, "zulip-tornado:*"])

    # Finally, restart the Django uWSGI processes.
    if (
        action == "restart"
        and not args.less_graceful
        and get_config(config_file, "application_server", "rolling_restart") != ""
        and os.path.exists("/home/zulip/deployments/uwsgi-control")
    ):
        # See if it's currently running
        uwsgi_status = subprocess.run(
            ["supervisorctl", "status", "zulip-django"],
            stdout=subprocess.DEVNULL,
        )
        if uwsgi_status.returncode == 0:
            logging.info("Starting rolling restart of django server")
            with open("/home/zulip/deployments/uwsgi-control", "w") as control_socket:
                # "c" is chain-reloading:
                # https://uwsgi-docs.readthedocs.io/en/latest/MasterFIFO.html#available-commands
                control_socket.write("c")
        else:
            logging.info("Starting django server")
            subprocess.check_call(["supervisorctl", "start", "zulip-django"])
    else:
        logging.info("%s django server", verbing)
        subprocess.check_call(["supervisorctl", action, "zulip-django"])

    using_sso = subprocess.check_output(["./scripts/get-django-setting", "USING_APACHE_SSO"])
    if using_sso.strip() == b"True":
        logging.info("Restarting Apache WSGI process...")
        subprocess.check_call(["pkill", "-x", "apache2", "-u", "zulip"])

# If we were doing this non-gracefully, or starting as opposed to
# restarting, we need to turn the workers (back) on.  There's no
# advantage to doing this not-all-at-once.
if (action == "start" or args.less_graceful) and len(workers) > 0:
    logging.info("Starting workers")
    subprocess.check_call(["supervisorctl", "start", *workers])

logging.info("Done!")
print(OKGREEN + f"Zulip {action}ed successfully!" + ENDC)

if change_symlink and "PWD" in os.environ:
    for symlink in [last_symlink, current_symlink]:
        if os.path.commonprefix([os.environ["PWD"], symlink]) == symlink:
            print(
                """
{}Your shell entered its current directory through a symlink:
  {}
which has now changed. Your shell will not see this change until you run:
  cd {}
to traverse the symlink again.{}
""".format(
                    WARNING, symlink, shlex.quote(os.environ["PWD"]), ENDC
                ),
                file=sys.stderr,
            )