scylla_swap_setup: Remove Before=swap.target dependency from swap unit
When a Scylla node starts, the scylla-image-setup.service invokes the
`scylla_swap_setup` script to provision swap. This script allocates a
swap file and creates a swap systemd unit to delegate control to
systemd. By default, systemd injects a Before=swap.target dependency
into every swap unit, allowing other services to use swap.target to wait
for swap to be enabled.
On Azure, this doesn't work so well because we store the swap file on
the ephemeral disk [1] which has network dependencies (`_netdev` mount
option, configured by cloud-init [2]). This makes the swap.target
indirectly depend on the network, leading to dependency cycles such as:
swap.target -> mnt-swapfile.swap -> mnt.mount -> network-online.target
-> network.target -> systemd-resolved.service -> tmp.mount -> swap.target
This patch breaks the cycle by removing the swap unit from swap.target
using DefaultDependencies=no. The swap unit will still be activated via
WantedBy=multi-user.target, just not during early boot.
Although this problem is specific to Azure, this patch applies the fix
to all clouds to keep the code simple.
Fixes #26519.
Fixes SCYLLADB-1257
[1] https://github.com/scylladb/scylla-machine-image/pull/426
[2] https://github.com/canonical/cloud-init/pull/1213#issuecomment-1026065501
Signed-off-by: Nikos Dragazis <nikolaos.dragazis@scylladb.com>
Closes scylladb/scylladb#28504
(cherry picked from commit 6d50e67bd2)
Closes scylladb/scylladb#29339
Closes scylladb/scylladb#29354
This commit is contained in:
committed by
Botond Dénes
parent
7b97fe4a92
commit
7ed772866e
28
dist/common/scripts/scylla_swap_setup
vendored
28
dist/common/scripts/scylla_swap_setup
vendored
@@ -9,6 +9,7 @@
|
||||
|
||||
import os
|
||||
import sys
|
||||
import shlex
|
||||
import argparse
|
||||
import psutil
|
||||
from pathlib import Path
|
||||
@@ -103,16 +104,41 @@ if __name__ == '__main__':
|
||||
run('dd if=/dev/zero of={} bs=1M count={}'.format(swapfile, swapsize_mb), shell=True, check=True)
|
||||
swapfile.chmod(0o600)
|
||||
run('mkswap -f {}'.format(swapfile), shell=True, check=True)
|
||||
|
||||
mount_point = find_mount_point(swap_directory)
|
||||
mount_unit = out(f'systemd-escape -p --suffix=mount {shlex.quote(str(mount_point))}')
|
||||
|
||||
# Add DefaultDependencies=no to the swap unit to avoid getting the default
|
||||
# Before=swap.target dependency. We apply this to all clouds, but the
|
||||
# requirement came from Azure:
|
||||
#
|
||||
# On Azure, the swap directory is on the Azure ephemeral disk (mounted on /mnt).
|
||||
# However, cloud-init makes this mount (i.e., the mnt.mount unit) depend on
|
||||
# the network (After=network-online.target). By extension, this means that
|
||||
# the swap unit depends on the network. If we didn't use DefaultDependencies=no,
|
||||
# then the swap unit would be part of the swap.target which other services
|
||||
# assume to be a local boot target, so we would end up with dependency cycles
|
||||
# such as:
|
||||
#
|
||||
# swap.target -> mnt-swapfile.swap -> mnt.mount -> network-online.target -> network.target -> systemd-resolved.service -> tmp.mount -> swap.target
|
||||
#
|
||||
# By removing the automatic Before=swap.target, the swap unit is no longer
|
||||
# part of swap.target, avoiding such cycles. The swap will still be
|
||||
# activated via WantedBy=multi-user.target.
|
||||
unit_data = '''
|
||||
[Unit]
|
||||
Description=swapfile
|
||||
DefaultDependencies=no
|
||||
After={}
|
||||
Conflicts=umount.target
|
||||
Before=umount.target
|
||||
|
||||
[Swap]
|
||||
What={}
|
||||
|
||||
[Install]
|
||||
WantedBy=multi-user.target
|
||||
'''[1:-1].format(swapfile)
|
||||
'''[1:-1].format(mount_unit, swapfile)
|
||||
with swapunit.open('w') as f:
|
||||
f.write(unit_data)
|
||||
systemd_unit.reload()
|
||||
|
||||
Reference in New Issue
Block a user