selfprivacy-rest-api/selfprivacy_api/backup/__init__.py

809 lines
26 KiB
Python
Raw Permalink Normal View History

2023-07-20 15:24:26 +00:00
"""
This module contains the controller class for backups.
"""
2024-07-26 19:59:44 +00:00
from datetime import datetime, timedelta, timezone
import time
import os
from os import statvfs
from typing import Callable, List, Optional
from os.path import exists
from selfprivacy_api.services import ServiceManager
from selfprivacy_api.services.service import (
Service,
ServiceStatus,
StoppedService,
)
from selfprivacy_api.jobs import Jobs, JobStatus, Job
2023-06-26 18:07:47 +00:00
2023-06-23 09:40:10 +00:00
from selfprivacy_api.graphql.queries.providers import (
BackupProvider as BackupProviderEnum,
)
from selfprivacy_api.graphql.common_types.backup import (
RestoreStrategy,
BackupReason,
AutobackupQuotas,
)
2023-06-26 18:07:47 +00:00
from selfprivacy_api.models.backup.snapshot import Snapshot
from selfprivacy_api.utils.block_devices import BlockDevices
2023-06-26 18:07:47 +00:00
2023-04-10 13:22:33 +00:00
from selfprivacy_api.backup.providers.provider import AbstractBackupProvider
from selfprivacy_api.backup.providers import get_provider
from selfprivacy_api.backup.storage import Storage
from selfprivacy_api.backup.jobs import (
get_backup_job,
get_backup_fail,
add_backup_job,
get_restore_job,
add_restore_job,
)
BACKUP_PROVIDER_ENVS = {
"kind": "BACKUP_KIND",
"login": "BACKUP_LOGIN",
"key": "BACKUP_KEY",
"location": "BACKUP_LOCATION",
}
AUTOBACKUP_JOB_EXPIRATION_SECONDS = 60 * 60 # one hour
2023-03-10 14:14:41 +00:00
class NotDeadError(AssertionError):
2023-07-20 15:24:26 +00:00
"""
This error is raised when we try to back up a service that is not dead yet.
"""
2023-07-20 16:37:01 +00:00
def __init__(self, service: Service):
self.service_name = service.get_id()
2023-07-20 15:24:26 +00:00
super().__init__()
def __str__(self):
return f"""
Service {self.service_name} should be either stopped or dead from
an error before we back up.
Normally, this error is unreachable because we do try ensure this.
Apparently, not this time.
"""
class RotationBucket:
"""
Bucket object used for rotation.
Has the following mutable fields:
- the counter, int
- the lambda function which takes datetime and the int and returns the int
- the last, int
"""
def __init__(self, counter: int, last: int, rotation_lambda):
self.counter: int = counter
self.last: int = last
self.rotation_lambda: Callable[[datetime, int], int] = rotation_lambda
def __str__(self) -> str:
return f"Bucket(counter={self.counter}, last={self.last})"
class Backups:
"""A stateless controller class for backups"""
# Providers
@staticmethod
def provider() -> AbstractBackupProvider:
2023-07-20 15:24:26 +00:00
"""
Returns the current backup storage provider.
"""
return Backups._lookup_provider()
@staticmethod
2023-06-23 09:40:10 +00:00
def set_provider(
kind: BackupProviderEnum,
login: str,
key: str,
location: str,
repo_id: str = "",
) -> None:
2023-07-20 15:24:26 +00:00
"""
Sets the new configuration of the backup storage provider.
In case of `BackupProviderEnum.BACKBLAZE`, the `login` is the key ID,
the `key` is the key itself, and the `location` is the bucket name and
the `repo_id` is the bucket ID.
"""
provider: AbstractBackupProvider = Backups._construct_provider(
2023-06-23 09:40:10 +00:00
kind,
login,
key,
location,
repo_id,
)
2023-04-10 13:22:33 +00:00
Storage.store_provider(provider)
@staticmethod
def reset() -> None:
2023-07-20 15:24:26 +00:00
"""
Deletes all the data about the backup storage provider.
"""
Storage.reset()
@staticmethod
def _lookup_provider() -> AbstractBackupProvider:
redis_provider = Backups._load_provider_redis()
if redis_provider is not None:
2023-03-10 14:14:41 +00:00
return redis_provider
none_provider = Backups._construct_provider(
2023-06-23 09:40:10 +00:00
BackupProviderEnum.NONE, login="", key="", location=""
)
Storage.store_provider(none_provider)
return none_provider
@staticmethod
def set_provider_from_envs():
for env in BACKUP_PROVIDER_ENVS.values():
if env not in os.environ.keys():
raise ValueError(
f"Cannot set backup provider from envs, there is no {env} set"
)
kind_str = os.environ[BACKUP_PROVIDER_ENVS["kind"]]
kind_enum = BackupProviderEnum[kind_str]
provider = Backups._construct_provider(
kind=kind_enum,
login=os.environ[BACKUP_PROVIDER_ENVS["login"]],
key=os.environ[BACKUP_PROVIDER_ENVS["key"]],
location=os.environ[BACKUP_PROVIDER_ENVS["location"]],
)
Storage.store_provider(provider)
@staticmethod
def _construct_provider(
kind: BackupProviderEnum,
login: str,
key: str,
location: str,
repo_id: str = "",
) -> AbstractBackupProvider:
provider_class = get_provider(kind)
return provider_class(
login=login,
key=key,
location=location,
repo_id=repo_id,
)
@staticmethod
def _load_provider_redis() -> Optional[AbstractBackupProvider]:
provider_model = Storage.load_provider()
if provider_model is None:
return None
return Backups._construct_provider(
BackupProviderEnum[provider_model.kind],
provider_model.login,
provider_model.key,
provider_model.location,
provider_model.repo_id,
)
# Init
@staticmethod
def init_repo() -> None:
2023-07-20 15:24:26 +00:00
"""
Initializes the backup repository. This is required once per repo.
"""
Backups.provider().backupper.init()
Storage.mark_as_init()
@staticmethod
def erase_repo() -> None:
"""
Completely empties the remote
"""
Backups.provider().backupper.erase_repo()
Storage.mark_as_uninitted()
@staticmethod
def is_initted() -> bool:
2023-07-20 15:24:26 +00:00
"""
Returns whether the backup repository is initialized or not.
If it is not initialized, we cannot back up and probably should
call `init_repo` first.
"""
if Storage.has_init_mark():
return True
initted = Backups.provider().backupper.is_initted()
if initted:
Storage.mark_as_init()
return True
return False
# Backup
2023-04-10 13:22:33 +00:00
@staticmethod
def back_up(
service: Service, reason: BackupReason = BackupReason.EXPLICIT
) -> Snapshot:
"""The top-level function to back up a service
If it fails for any reason at all, it should both mark job as
errored and re-raise an error"""
2023-02-08 15:27:49 +00:00
job = get_backup_job(service)
if job is None:
job = add_backup_job(service)
Jobs.update(job, status=JobStatus.RUNNING)
try:
if service.can_be_backed_up() is False:
raise ValueError("cannot backup a non-backuppable service")
folders = service.get_folders()
service_name = service.get_id()
service.pre_backup()
snapshot = Backups.provider().backupper.start_backup(
folders,
service_name,
reason=reason,
)
2023-08-21 12:45:31 +00:00
Backups._on_new_snapshot_created(service_name, snapshot)
2023-08-21 12:45:31 +00:00
if reason == BackupReason.AUTO:
Backups._prune_auto_snaps(service)
service.post_restore()
2023-07-20 15:24:26 +00:00
except Exception as error:
Jobs.update(job, status=JobStatus.ERROR, error=str(error))
2023-07-20 15:24:26 +00:00
raise error
Jobs.update(job, status=JobStatus.FINISHED)
if reason in [BackupReason.AUTO, BackupReason.PRE_RESTORE]:
Jobs.set_expiration(job, AUTOBACKUP_JOB_EXPIRATION_SECONDS)
return Backups.sync_date_from_cache(snapshot)
@staticmethod
def sync_date_from_cache(snapshot: Snapshot) -> Snapshot:
"""
Our snapshot creation dates are different from those on server by a tiny amount.
This is a convenience, maybe it is better to write a special comparison
function for snapshots
"""
snap = Storage.get_cached_snapshot_by_id(snapshot.id)
if snap is None:
raise ValueError(
f"snapshot {snapshot.id} date syncing failed, this should never happen normally"
)
return snap
2023-08-21 12:45:31 +00:00
@staticmethod
def _auto_snaps(service) -> List[Snapshot]:
2023-08-21 12:45:31 +00:00
return [
snap
for snap in Backups.get_snapshots(service)
if snap.reason == BackupReason.AUTO
]
@staticmethod
def _prune_snaps_with_quotas(snapshots: List[Snapshot]) -> List[Snapshot]:
# Function broken out for testability
# Sorting newest first
sorted_snaps = sorted(snapshots, key=lambda s: s.created_at, reverse=True)
quotas: AutobackupQuotas = Backups.autobackup_quotas()
buckets: list[RotationBucket] = [
RotationBucket(
quotas.last, # type: ignore
-1,
lambda _, index: index,
),
RotationBucket(
quotas.daily, # type: ignore
-1,
lambda date, _: date.year * 10000 + date.month * 100 + date.day,
),
RotationBucket(
quotas.weekly, # type: ignore
-1,
lambda date, _: date.year * 100 + date.isocalendar()[1],
),
RotationBucket(
quotas.monthly, # type: ignore
-1,
lambda date, _: date.year * 100 + date.month,
),
RotationBucket(
quotas.yearly, # type: ignore
-1,
lambda date, _: date.year,
),
]
new_snaplist: List[Snapshot] = []
for i, snap in enumerate(sorted_snaps):
keep_snap = False
for bucket in buckets:
if (bucket.counter > 0) or (bucket.counter == -1):
val = bucket.rotation_lambda(snap.created_at, i)
if (val != bucket.last) or (i == len(sorted_snaps) - 1):
bucket.last = val
if bucket.counter > 0:
bucket.counter -= 1
if not keep_snap:
new_snaplist.append(snap)
keep_snap = True
return new_snaplist
2023-08-21 12:45:31 +00:00
@staticmethod
def _prune_auto_snaps(service) -> None:
# Not very testable by itself, so most testing is going on Backups._prune_snaps_with_quotas
# We can still test total limits and, say, daily limits
2023-08-21 12:45:31 +00:00
auto_snaps = Backups._auto_snaps(service)
new_snaplist = Backups._prune_snaps_with_quotas(auto_snaps)
deletable_snaps = [snap for snap in auto_snaps if snap not in new_snaplist]
Backups.forget_snapshots(deletable_snaps)
2023-08-21 12:45:31 +00:00
@staticmethod
def _standardize_quotas(i: int) -> int:
if i <= -1:
i = -1
return i
@staticmethod
def autobackup_quotas() -> AutobackupQuotas:
"""0 means do not keep, -1 means unlimited"""
return Storage.autobackup_quotas()
@staticmethod
def set_autobackup_quotas(quotas: AutobackupQuotas) -> None:
"""0 means do not keep, -1 means unlimited"""
Storage.set_autobackup_quotas(
AutobackupQuotas(
last=Backups._standardize_quotas(quotas.last), # type: ignore
daily=Backups._standardize_quotas(quotas.daily), # type: ignore
weekly=Backups._standardize_quotas(quotas.weekly), # type: ignore
monthly=Backups._standardize_quotas(quotas.monthly), # type: ignore
yearly=Backups._standardize_quotas(quotas.yearly), # type: ignore
)
)
# do not prune all autosnaps right away, this will be done by an async task
@staticmethod
def prune_all_autosnaps() -> None:
for service in ServiceManager.get_all_services():
Backups._prune_auto_snaps(service)
# Restoring
@staticmethod
def _ensure_queued_restore_job(service, snapshot) -> Job:
job = get_restore_job(service)
if job is None:
job = add_restore_job(snapshot)
Jobs.update(job, status=JobStatus.CREATED)
return job
2023-03-14 00:39:15 +00:00
@staticmethod
def _inplace_restore(
service: Service,
snapshot: Snapshot,
job: Job,
) -> None:
Jobs.update(
job, status=JobStatus.CREATED, status_text="Waiting for pre-restore backup"
)
failsafe_snapshot = Backups.back_up(service, BackupReason.PRE_RESTORE)
Jobs.update(
job, status=JobStatus.RUNNING, status_text=f"Restoring from {snapshot.id}"
)
try:
Backups._restore_service_from_snapshot(
service,
snapshot.id,
verify=False,
)
2023-07-20 15:24:26 +00:00
except Exception as error:
Jobs.update(
job,
status=JobStatus.ERROR,
status_text=f"Restore failed with {str(error)}, reverting to {failsafe_snapshot.id}",
)
Backups._restore_service_from_snapshot(
service, failsafe_snapshot.id, verify=False
)
Jobs.update(
job,
status=JobStatus.ERROR,
status_text=f"Restore failed with {str(error)}, reverted to {failsafe_snapshot.id}",
)
2023-07-20 15:24:26 +00:00
raise error
@staticmethod
def restore_snapshot(
snapshot: Snapshot, strategy=RestoreStrategy.DOWNLOAD_VERIFY_OVERWRITE
) -> None:
2023-07-20 15:24:26 +00:00
"""Restores a snapshot to its original service using the given strategy"""
service = ServiceManager.get_service_by_id(snapshot.service_name)
if service is None:
raise ValueError(
f"snapshot has a nonexistent service: {snapshot.service_name}"
)
job = Backups._ensure_queued_restore_job(service, snapshot)
try:
Backups._assert_restorable(snapshot)
Jobs.update(
job, status=JobStatus.RUNNING, status_text="Stopping the service"
)
with StoppedService(service):
if not service.is_always_active():
Backups.assert_dead(service)
if strategy == RestoreStrategy.INPLACE:
Backups._inplace_restore(service, snapshot, job)
else: # verify_before_download is our default
Jobs.update(
job,
status=JobStatus.RUNNING,
status_text=f"Restoring from {snapshot.id}",
)
Backups._restore_service_from_snapshot(
service, snapshot.id, verify=True
)
service.post_restore()
Jobs.update(
job,
status=JobStatus.RUNNING,
progress=90,
status_text="Restarting the service",
)
2023-07-20 15:24:26 +00:00
except Exception as error:
Jobs.update(job, status=JobStatus.ERROR, status_text=str(error))
2023-07-20 15:24:26 +00:00
raise error
Jobs.update(job, status=JobStatus.FINISHED)
@staticmethod
def _assert_restorable(
snapshot: Snapshot, strategy=RestoreStrategy.DOWNLOAD_VERIFY_OVERWRITE
) -> None:
service = ServiceManager.get_service_by_id(snapshot.service_name)
if service is None:
raise ValueError(
f"snapshot has a nonexistent service: {snapshot.service_name}"
)
restored_snap_size = Backups.snapshot_restored_size(snapshot.id)
if strategy == RestoreStrategy.DOWNLOAD_VERIFY_OVERWRITE:
needed_space = restored_snap_size
elif strategy == RestoreStrategy.INPLACE:
needed_space = restored_snap_size - service.get_storage_usage()
else:
raise NotImplementedError(
"""
We do not know if there is enough space for restoration because
there is some novel restore strategy used!
This is a developer's fault, open an issue please
"""
)
available_space = Backups.space_usable_for_service(service)
if needed_space > available_space:
raise ValueError(
f"we only have {available_space} bytes "
f"but snapshot needs {needed_space}"
)
@staticmethod
def _restore_service_from_snapshot(
service: Service,
snapshot_id: str,
verify=True,
) -> None:
folders = service.get_folders()
Backups.provider().backupper.restore_from_backup(
snapshot_id,
folders,
verify=verify,
)
# Snapshots
@staticmethod
def get_snapshots(service: Service) -> List[Snapshot]:
2023-07-20 15:24:26 +00:00
"""Returns all snapshots for a given service"""
snapshots = Backups.get_all_snapshots()
2023-06-23 09:40:10 +00:00
service_id = service.get_id()
return list(
filter(
lambda snap: snap.service_name == service_id,
snapshots,
)
)
@staticmethod
def get_all_snapshots() -> List[Snapshot]:
2023-07-20 15:24:26 +00:00
"""Returns all snapshots"""
# When we refresh our cache:
# 1. Manually
# 2. On timer
# 3. On new snapshot
# 4. On snapshot deletion
return Storage.get_cached_snapshots()
@staticmethod
2023-07-20 15:24:26 +00:00
def get_snapshot_by_id(snapshot_id: str) -> Optional[Snapshot]:
"""Returns a backup snapshot by its id"""
snap = Storage.get_cached_snapshot_by_id(snapshot_id)
if snap is not None:
return snap
# Possibly our cache entry got invalidated, let's try one more time
Backups.force_snapshot_cache_reload()
2023-07-20 15:24:26 +00:00
snap = Storage.get_cached_snapshot_by_id(snapshot_id)
return snap
@staticmethod
def forget_snapshots(snapshots: List[Snapshot]) -> None:
"""
Deletes a batch of snapshots from the repo and syncs cache
Optimized
"""
ids = [snapshot.id for snapshot in snapshots]
Backups.provider().backupper.forget_snapshots(ids)
Backups.force_snapshot_cache_reload()
2023-07-05 13:13:30 +00:00
@staticmethod
def forget_snapshot(snapshot: Snapshot) -> None:
"""Deletes a snapshot from the repo and from cache"""
Backups.forget_snapshots([snapshot])
2023-07-05 13:13:30 +00:00
@staticmethod
def forget_all_snapshots():
"""
Mark all snapshots we have made for deletion and make them inaccessible
(this is done by cloud, we only issue a command)
"""
Backups.forget_snapshots(Backups.get_all_snapshots())
@staticmethod
def force_snapshot_cache_reload() -> None:
2023-07-20 15:24:26 +00:00
"""
Forces a reload of the snapshot cache.
This may be an expensive operation, so use it wisely.
User pays for the API calls.
"""
upstream_snapshots = Backups.provider().backupper.get_snapshots()
Storage.invalidate_snapshot_storage()
for snapshot in upstream_snapshots:
Storage.cache_snapshot(snapshot)
@staticmethod
def snapshot_restored_size(snapshot_id: str) -> int:
2023-07-20 15:24:26 +00:00
"""Returns the size of the snapshot"""
return Backups.provider().backupper.restored_size(
snapshot_id,
)
@staticmethod
def _on_new_snapshot_created(service_id: str, snapshot: Snapshot) -> None:
"""What do we do with a snapshot that is just made?"""
# non-expiring timestamp of the last
Storage.store_last_timestamp(service_id, snapshot)
Backups.force_snapshot_cache_reload()
# Autobackup
@staticmethod
def autobackup_period_minutes() -> Optional[int]:
"""None means autobackup is disabled"""
return Storage.autobackup_period_minutes()
@staticmethod
def set_autobackup_period_minutes(minutes: int) -> None:
"""
0 and negative numbers are equivalent to disable.
Setting to a positive number may result in a backup very soon
if some services are not backed up.
"""
if minutes <= 0:
Backups.disable_all_autobackup()
return
Storage.store_autobackup_period_minutes(minutes)
@staticmethod
def disable_all_autobackup() -> None:
"""
Disables all automatic backing up,
but does not change per-service settings
"""
Storage.delete_backup_period()
@staticmethod
def is_time_to_backup(time: datetime) -> bool:
"""
Intended as a time validator for huey cron scheduler
of automatic backups
"""
return Backups.services_to_back_up(time) != []
@staticmethod
def services_to_back_up(time: datetime) -> List[Service]:
2023-07-20 15:24:26 +00:00
"""Returns a list of services that should be backed up at a given time"""
return [
service
for service in ServiceManager.get_all_services()
if Backups.is_time_to_backup_service(service, time)
]
@staticmethod
def get_last_backed_up(service: Service) -> Optional[datetime]:
"""Get a timezone-aware time of the last backup of a service"""
return Storage.get_last_backup_time(service.get_id())
@staticmethod
def get_last_backup_error_time(service: Service) -> Optional[datetime]:
"""Get a timezone-aware time of the last backup of a service"""
job = get_backup_fail(service)
if job is not None:
datetime_created = job.created_at
if datetime_created.tzinfo is None:
# assume it is in localtime
offset = timedelta(seconds=time.localtime().tm_gmtoff)
datetime_created = datetime_created - offset
2023-11-13 16:36:12 +00:00
return datetime.combine(
datetime_created.date(), datetime_created.time(), timezone.utc
)
return datetime_created
return None
@staticmethod
def is_time_to_backup_service(service: Service, time: datetime):
2023-07-20 15:24:26 +00:00
"""Returns True if it is time to back up a service"""
period = Backups.autobackup_period_minutes()
if period is None:
return False
if not service.is_enabled():
return False
if not service.can_be_backed_up():
return False
last_error = Backups.get_last_backup_error_time(service)
if last_error is not None:
if time < last_error + timedelta(seconds=AUTOBACKUP_JOB_EXPIRATION_SECONDS):
return False
last_backup = Backups.get_last_backed_up(service)
# Queue a backup immediately if there are no previous backups
if last_backup is None:
return True
if time > last_backup + timedelta(minutes=period):
return True
return False
# Helpers
2023-04-10 13:22:33 +00:00
@staticmethod
2023-06-23 09:40:10 +00:00
def space_usable_for_service(service: Service) -> int:
2023-07-20 15:24:26 +00:00
"""
Returns the amount of space available on the volume the given
service is located on.
"""
folders = service.get_folders()
if folders == []:
raise ValueError("unallocated service", service.get_id())
# We assume all folders of one service live at the same volume
example_folder = folders[0]
if exists(example_folder):
fs_info = statvfs(example_folder)
usable_bytes = fs_info.f_frsize * fs_info.f_bavail
else:
# Look at the block device as it is written in settings
label = service.get_drive()
device = BlockDevices().get_block_device(label)
if device is None:
raise ValueError("nonexistent drive ", label, " for ", service.get_id())
usable_bytes = int(device.fsavail)
return usable_bytes
2023-04-10 13:22:33 +00:00
@staticmethod
def set_localfile_repo(file_path: str):
2023-07-20 15:24:26 +00:00
"""Used by tests to set a local folder as a backup repo"""
# pylint: disable-next=invalid-name
ProviderClass = get_provider(BackupProviderEnum.FILE)
provider = ProviderClass(
login="",
key="",
location=file_path,
repo_id="",
)
Storage.store_provider(provider)
@staticmethod
def assert_dead(service: Service):
"""
2023-07-20 15:24:26 +00:00
Checks if a service is dead and can be safely restored from a snapshot.
"""
if service.get_status() not in [
ServiceStatus.INACTIVE,
ServiceStatus.FAILED,
]:
raise NotDeadError(service)
@staticmethod
def is_same_slice(snap1: Snapshot, snap2: Snapshot) -> bool:
# Determines if the snaps were made roughly in the same time period
period_minutes = Backups.autobackup_period_minutes()
# Autobackups are not guaranteed to be enabled during restore.
# If they are not, period will be none
# We ASSUME that picking latest snap of the same day is safe enough
# But it is potentlially problematic and is better done with metadata I think.
if period_minutes is None:
period_minutes = 24 * 60
if snap1.created_at > snap2.created_at + timedelta(minutes=period_minutes):
return False
if snap1.created_at < snap2.created_at - timedelta(minutes=period_minutes):
return False
return True
@staticmethod
def last_backup_slice() -> List[Snapshot]:
"""
Guarantees that the slice is valid, ie, it has an api snapshot too
Or empty
"""
slice: List[Snapshot] = []
# We need snapshots that were made around the same time.
# And we need to be sure that api snap is in there
# That's why we form the slice around api snap
api_snaps = Backups.get_snapshots(ServiceManager())
if api_snaps == []:
return []
api_snaps.sort(key=lambda x: x.created_at, reverse=True)
api_snap = api_snaps[0] # pick the latest one
for service in ServiceManager.get_all_services():
if isinstance(service, ServiceManager):
continue
snaps = Backups.get_snapshots(service)
snaps.sort(key=lambda x: x.created_at, reverse=True)
for snap in snaps:
if Backups.is_same_slice(snap, api_snap):
slice.append(snap)
break
slice.append(api_snap)
return slice