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

591 lines
18 KiB
Python

"""
This module contains the controller class for backups.
"""
from datetime import datetime, timedelta
import os
from os import statvfs
from typing import List, Optional
from selfprivacy_api.utils import ReadUserData, WriteUserData
from selfprivacy_api.services import (
get_service_by_id,
get_all_services,
)
from selfprivacy_api.services.service import (
Service,
ServiceStatus,
StoppedService,
)
from selfprivacy_api.jobs import Jobs, JobStatus, Job
from selfprivacy_api.graphql.queries.providers import (
BackupProvider as BackupProviderEnum,
)
from selfprivacy_api.graphql.common_types.backup import RestoreStrategy
from selfprivacy_api.models.backup.snapshot import Snapshot
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,
add_backup_job,
get_restore_job,
add_restore_job,
)
DEFAULT_JSON_PROVIDER = {
"provider": "BACKBLAZE",
"accountId": "",
"accountKey": "",
"bucket": "",
}
BACKUP_PROVIDER_ENVS = {
"kind": "BACKUP_KIND",
"login": "BACKUP_LOGIN",
"key": "BACKUP_KEY",
"location": "BACKUP_LOCATION",
}
class NotDeadError(AssertionError):
"""
This error is raised when we try to back up a service that is not dead yet.
"""
def __init__(self, service: Service):
self.service_name = service.get_id()
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 Backups:
"""A stateless controller class for backups"""
# Providers
@staticmethod
def provider() -> AbstractBackupProvider:
"""
Returns the current backup storage provider.
"""
return Backups._lookup_provider()
@staticmethod
def set_provider(
kind: BackupProviderEnum,
login: str,
key: str,
location: str,
repo_id: str = "",
) -> None:
"""
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(
kind,
login,
key,
location,
repo_id,
)
Storage.store_provider(provider)
@staticmethod
def reset(reset_json=True) -> None:
"""
Deletes all the data about the backup storage provider.
"""
Storage.reset()
if reset_json:
try:
Backups._reset_provider_json()
except FileNotFoundError:
# if there is no userdata file, we do not need to reset it
pass
@staticmethod
def _lookup_provider() -> AbstractBackupProvider:
redis_provider = Backups._load_provider_redis()
if redis_provider is not None:
return redis_provider
try:
json_provider = Backups._load_provider_json()
except FileNotFoundError:
json_provider = None
if json_provider is not None:
Storage.store_provider(json_provider)
return json_provider
none_provider = Backups._construct_provider(
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,
)
@staticmethod
def _load_provider_json() -> Optional[AbstractBackupProvider]:
with ReadUserData() as user_data:
provider_dict = {
"provider": "",
"accountId": "",
"accountKey": "",
"bucket": "",
}
if "backup" not in user_data.keys():
if "backblaze" in user_data.keys():
provider_dict.update(user_data["backblaze"])
provider_dict["provider"] = "BACKBLAZE"
return None
else:
provider_dict.update(user_data["backup"])
if provider_dict == DEFAULT_JSON_PROVIDER:
return None
try:
return Backups._construct_provider(
kind=BackupProviderEnum[provider_dict["provider"]],
login=provider_dict["accountId"],
key=provider_dict["accountKey"],
location=provider_dict["bucket"],
)
except KeyError:
return None
@staticmethod
def _reset_provider_json() -> None:
with WriteUserData() as user_data:
if "backblaze" in user_data.keys():
del user_data["backblaze"]
user_data["backup"] = DEFAULT_JSON_PROVIDER
# Init
@staticmethod
def init_repo() -> None:
"""
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:
"""
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
@staticmethod
def back_up(service: Service) -> Snapshot:
"""The top-level function to back up a service"""
folders = service.get_folders()
tag = service.get_id()
job = get_backup_job(service)
if job is None:
job = add_backup_job(service)
Jobs.update(job, status=JobStatus.RUNNING)
try:
service.pre_backup()
snapshot = Backups.provider().backupper.start_backup(
folders,
tag,
)
Backups._store_last_snapshot(tag, snapshot)
service.post_restore()
except Exception as error:
Jobs.update(job, status=JobStatus.ERROR)
raise error
Jobs.update(job, status=JobStatus.FINISHED)
return snapshot
# 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
@staticmethod
def _inplace_restore(
service: Service,
snapshot: Snapshot,
job: Job,
) -> None:
failsafe_snapshot = Backups.back_up(service)
Jobs.update(job, status=JobStatus.RUNNING)
try:
Backups._restore_service_from_snapshot(
service,
snapshot.id,
verify=False,
)
except Exception as error:
Backups._restore_service_from_snapshot(
service, failsafe_snapshot.id, verify=False
)
raise error
@staticmethod
def restore_snapshot(
snapshot: Snapshot, strategy=RestoreStrategy.DOWNLOAD_VERIFY_OVERWRITE
) -> None:
"""Restores a snapshot to its original service using the given strategy"""
service = 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)
with StoppedService(service):
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)
Backups._restore_service_from_snapshot(
service, snapshot.id, verify=True
)
service.post_restore()
except Exception as error:
Jobs.update(job, status=JobStatus.ERROR)
raise error
Jobs.update(job, status=JobStatus.FINISHED)
@staticmethod
def _assert_restorable(
snapshot: Snapshot, strategy=RestoreStrategy.DOWNLOAD_VERIFY_OVERWRITE
) -> None:
service = 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]:
"""Returns all snapshots for a given service"""
snapshots = Backups.get_all_snapshots()
service_id = service.get_id()
return list(
filter(
lambda snap: snap.service_name == service_id,
snapshots,
)
)
@staticmethod
def get_all_snapshots() -> List[Snapshot]:
"""Returns all snapshots"""
cached_snapshots = Storage.get_cached_snapshots()
if cached_snapshots:
return cached_snapshots
# TODO: the oldest snapshots will get expired faster than the new ones.
# How to detect that the end is missing?
Backups.force_snapshot_cache_reload()
return Storage.get_cached_snapshots()
@staticmethod
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()
snap = Storage.get_cached_snapshot_by_id(snapshot_id)
return snap
@staticmethod
def forget_snapshot(snapshot: Snapshot) -> None:
"""Deletes a snapshot from the repo and from cache"""
Backups.provider().backupper.forget_snapshot(snapshot.id)
Storage.delete_cached_snapshot(snapshot)
@staticmethod
def forget_all_snapshots():
"""deliberately erase all snapshots we made"""
# there is no dedicated optimized command for this,
# but maybe we can have a multi-erase
for snapshot in Backups.get_all_snapshots():
Backups.forget_snapshot(snapshot)
@staticmethod
def force_snapshot_cache_reload() -> None:
"""
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:
"""Returns the size of the snapshot"""
return Backups.provider().backupper.restored_size(
snapshot_id,
)
@staticmethod
def _store_last_snapshot(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)
# expiring cache entry
Storage.cache_snapshot(snapshot)
# 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]:
"""Returns a list of services that should be backed up at a given time"""
return [
service
for service in 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 is_time_to_backup_service(service: Service, time: datetime):
"""Returns True if it is time to back up a service"""
period = Backups.autobackup_period_minutes()
service_id = service.get_id()
if not service.can_be_backed_up():
return False
if period is None:
return False
last_backup = Storage.get_last_backup_time(service_id)
if last_backup is None:
# queue a backup immediately if there are no previous backups
return True
if time > last_backup + timedelta(minutes=period):
return True
return False
# Helpers
@staticmethod
def space_usable_for_service(service: Service) -> int:
"""
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
fs_info = statvfs(folders[0])
usable_bytes = fs_info.f_frsize * fs_info.f_bavail
return usable_bytes
@staticmethod
def set_localfile_repo(file_path: str):
"""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):
"""
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)