selfprivacy-rest-api/tests/test_backup.py

import pytest

import os
import os.path as path
from os import remove
from os import listdir
from os import urandom

from datetime import datetime, timedelta, timezone
import tempfile

from selfprivacy_api.utils.huey import huey


from selfprivacy_api.services.service import ServiceStatus
from selfprivacy_api.services import ServiceManager

from selfprivacy_api.graphql.queries.providers import BackupProvider as ProviderEnum
from selfprivacy_api.graphql.common_types.backup import (
    RestoreStrategy,
    BackupReason,
)
from selfprivacy_api.graphql.queries.providers import BackupProvider

from selfprivacy_api.jobs import Job, Jobs, JobStatus

from selfprivacy_api.models.backup.snapshot import Snapshot

from selfprivacy_api.backup import Backups, BACKUP_PROVIDER_ENVS
import selfprivacy_api.backup.providers as providers
from selfprivacy_api.backup.providers import AbstractBackupProvider
from selfprivacy_api.backup.providers.backblaze import Backblaze
from selfprivacy_api.backup.providers.none import NoBackups
from selfprivacy_api.backup.providers import get_kind
from selfprivacy_api.backup.util import sync

from selfprivacy_api.backup.tasks import (
    start_backup,
    restore_snapshot,
    reload_snapshot_cache,
)
from selfprivacy_api.backup.storage import Storage
from selfprivacy_api.backup.local_secret import LocalBackupSecret
from selfprivacy_api.backup.jobs import get_backup_fail

from tests.common import assert_job_errored
from tests.test_dkim import dkim_file


REPO_NAME = "test_backup"

REPOFILE_NAME = "totallyunrelated"


def prepare_localfile_backups(temp_dir):
    test_repo_path = path.join(temp_dir, REPOFILE_NAME)
    assert not path.exists(test_repo_path)
    Backups.set_localfile_repo(test_repo_path)


@pytest.fixture(scope="function")
def backups_local(tmpdir):
    Backups.reset()
    prepare_localfile_backups(tmpdir)
    Jobs.reset()
    Backups.init_repo()


@pytest.fixture(scope="function")
def backups(tmpdir):
    """
    For those tests that are supposed to pass with
    both local and cloud repos
    """

    # Sometimes this is false. Idk why.
    huey.immediate = True
    assert huey.immediate is True

    Backups.reset()
    if BACKUP_PROVIDER_ENVS["kind"] in os.environ.keys():
        Backups.set_provider_from_envs()
    else:
        prepare_localfile_backups(tmpdir)
    Jobs.reset()

    Backups.init_repo()
    assert Backups.provider().location == str(tmpdir) + "/" + REPOFILE_NAME
    yield
    Backups.erase_repo()


@pytest.fixture()
def memory_backup() -> AbstractBackupProvider:
    ProviderClass = providers.get_provider(BackupProvider.MEMORY)
    assert ProviderClass is not None
    memory_provider = ProviderClass(login="", key="")
    assert memory_provider is not None
    return memory_provider


@pytest.fixture()
def file_backup(tmpdir) -> AbstractBackupProvider:
    test_repo_path = path.join(tmpdir, "test_repo")
    ProviderClass = providers.get_provider(BackupProvider.FILE)
    assert ProviderClass is not None
    provider = ProviderClass(location=test_repo_path)
    assert provider is not None
    return provider


def test_reset_sets_to_none1():
    Backups.reset()
    provider = Backups.provider()
    assert provider is not None
    assert isinstance(provider, NoBackups)


def test_reset_sets_to_none2(backups):
    # now with something set up first^^^
    Backups.reset()
    provider = Backups.provider()
    assert provider is not None
    assert isinstance(provider, NoBackups)


def test_setting_from_envs(tmpdir):
    Backups.reset()
    environment_stash = {}
    if BACKUP_PROVIDER_ENVS["kind"] in os.environ.keys():
        # we are running under special envs, stash them before rewriting them
        for key in BACKUP_PROVIDER_ENVS.values():
            environment_stash[key] = os.environ[key]

    os.environ[BACKUP_PROVIDER_ENVS["kind"]] = "BACKBLAZE"
    os.environ[BACKUP_PROVIDER_ENVS["login"]] = "ID"
    os.environ[BACKUP_PROVIDER_ENVS["key"]] = "KEY"
    os.environ[BACKUP_PROVIDER_ENVS["location"]] = "selfprivacy"
    Backups.set_provider_from_envs()
    provider = Backups.provider()

    assert provider is not None
    assert isinstance(provider, Backblaze)
    assert provider.login == "ID"
    assert provider.key == "KEY"
    assert provider.location == "selfprivacy"

    assert provider.backupper.account == "ID"
    assert provider.backupper.key == "KEY"

    if environment_stash != {}:
        for key in BACKUP_PROVIDER_ENVS.values():
            os.environ[key] = environment_stash[key]
    else:
        for key in BACKUP_PROVIDER_ENVS.values():
            del os.environ[key]


def test_select_backend():
    provider = providers.get_provider(BackupProvider.BACKBLAZE)
    assert provider is not None
    assert provider == Backblaze


def test_file_backend_init(file_backup):
    file_backup.backupper.init()


def test_reinit_after_purge(backups):
    assert Backups.is_initted() is True

    Backups.erase_repo()
    assert Backups.is_initted() is False
    with pytest.raises(ValueError):
        Backups.force_snapshot_cache_reload()

    Backups.init_repo()
    assert Backups.is_initted() is True
    assert len(Backups.get_all_snapshots()) == 0


def test_backup_service(dummy_service, backups):
    id = dummy_service.get_id()
    assert_job_finished(f"services.{id}.backup", count=0)
    assert Backups.get_last_backed_up(dummy_service) is None

    Backups.back_up(dummy_service)

    now = datetime.now(timezone.utc)
    date = Backups.get_last_backed_up(dummy_service)
    assert date is not None
    assert now > date
    assert now - date < timedelta(minutes=1)

    assert_job_finished(f"services.{id}.backup", count=1)


def all_job_text(job: Job) -> str:
    # Use when we update to pydantic 2.xxx
    # return Job.model_dump_json()
    result = ""
    if job.status_text is not None:
        result += job.status_text
    if job.description is not None:
        result += job.description
    if job.error is not None:
        result += job.error

    return result


def test_error_censoring_encryptionkey(dummy_service, backups):
    # Discard our key to inject a failure
    old_key = LocalBackupSecret.get()
    LocalBackupSecret.reset()
    new_key = LocalBackupSecret.get()

    with pytest.raises(ValueError):
        # Should fail without correct key
        Backups.back_up(dummy_service)

    job = get_backup_fail(dummy_service)
    assert_job_errored(job)

    job_text = all_job_text(job)

    assert old_key not in job_text
    assert new_key not in job_text
    # local backups do not have login key
    # assert Backups.provider().key not in job_text

    assert "CENSORED" in job_text


def test_error_censoring_loginkey(dummy_service, backups, fp):
    # We do not want to screw up our teardown
    old_provider = Backups.provider()

    secret = "aSecretNYA"

    Backups.set_provider(
        ProviderEnum.BACKBLAZE, login="meow", key=secret, location="moon"
    )
    assert Backups.provider().key == secret

    # We could have called real backblaze but it is kind of not privacy so.
    fp.allow_unregistered(True)
    fp.register(
        ["restic", fp.any()],
        returncode=1,
        stdout="only real cats are allowed",
        # We do not want to suddenly call real backblaze even if code changes
        occurrences=100,
    )

    with pytest.raises(ValueError):
        Backups.back_up(dummy_service)

    job = get_backup_fail(dummy_service)
    assert_job_errored(job)

    job_text = all_job_text(job)
    assert secret not in job_text
    assert job_text.count("CENSORED") == 2

    # We do not want to screw up our teardown
    Storage.store_provider(old_provider)


def test_no_repo(memory_backup):
    with pytest.raises(ValueError):
        assert memory_backup.backupper.get_snapshots() == []


def test_one_snapshot(backups, dummy_service):
    Backups.back_up(dummy_service)

    snaps = Backups.get_snapshots(dummy_service)
    assert len(snaps) == 1
    snap = snaps[0]
    assert snap.service_name == dummy_service.get_id()


def test_backup_returns_snapshot(backups, dummy_service):
    service_folders = dummy_service.get_folders()
    provider = Backups.provider()
    name = dummy_service.get_id()
    snapshot = provider.backupper.start_backup(service_folders, name)

    assert snapshot.id is not None

    snapshots = provider.backupper.get_snapshots()
    assert snapshots != []

    assert len(snapshot.id) == len(snapshots[0].id)
    assert Backups.get_snapshot_by_id(snapshot.id) is not None
    assert snapshot.service_name == name
    assert snapshot.created_at is not None
    assert snapshot.reason == BackupReason.EXPLICIT


def test_backup_reasons(backups, dummy_service):
    snap = Backups.back_up(dummy_service, BackupReason.AUTO)
    assert snap.reason == BackupReason.AUTO

    Backups.force_snapshot_cache_reload()
    snaps = Backups.get_snapshots(dummy_service)
    assert snaps[0].reason == BackupReason.AUTO


def folder_files(folder):
    return [
        path.join(folder, filename)
        for filename in listdir(folder)
        if filename is not None
    ]


def service_files(service):
    result = []
    for service_folder in service.get_folders():
        result.extend(folder_files(service_folder))
    return result


def test_restore(backups, dummy_service):
    paths_to_nuke = service_files(dummy_service)
    contents = []

    for service_file in paths_to_nuke:
        with open(service_file, "r") as file:
            contents.append(file.read())

    Backups.back_up(dummy_service)
    snap = Backups.get_snapshots(dummy_service)[0]
    assert snap is not None

    for p in paths_to_nuke:
        assert path.exists(p)
        remove(p)
        assert not path.exists(p)

    Backups._restore_service_from_snapshot(dummy_service, snap.id)
    for p, content in zip(paths_to_nuke, contents):
        assert path.exists(p)
        with open(p, "r") as file:
            assert file.read() == content


def test_sizing(backups, dummy_service):
    Backups.back_up(dummy_service)
    snap = Backups.get_snapshots(dummy_service)[0]
    size = Backups.snapshot_restored_size(snap.id)
    assert size is not None
    assert size > 0


def test_init_tracking(backups, tmpdir):
    assert Backups.is_initted() is True
    Backups.reset()
    assert Backups.is_initted() is False
    separate_dir = tmpdir / "out_of_the_way"
    prepare_localfile_backups(separate_dir)
    Backups.init_repo()

    assert Backups.is_initted() is True


def finished_jobs():
    return [job for job in Jobs.get_jobs() if job.status is JobStatus.FINISHED]


def assert_job_finished(job_type, count):
    finished_types = [job.type_id for job in finished_jobs()]
    assert finished_types.count(job_type) == count


def assert_job_has_run(job_type):
    job = [job for job in finished_jobs() if job.type_id == job_type][0]
    assert JobStatus.RUNNING in Jobs.status_updates(job)


def job_progress_updates(job_type):
    job = [job for job in finished_jobs() if job.type_id == job_type][0]
    return Jobs.progress_updates(job)


def assert_job_had_progress(job_type):
    assert len(job_progress_updates(job_type)) > 0


def make_large_file(path: str, bytes: int):
    with open(path, "wb") as file:
        file.write(urandom(bytes))


def test_snapshots_by_id(backups, dummy_service):
    snap1 = Backups.back_up(dummy_service)
    snap2 = Backups.back_up(dummy_service)
    snap3 = Backups.back_up(dummy_service)

    assert snap2.id is not None
    assert snap2.id != ""

    assert len(Backups.get_snapshots(dummy_service)) == 3
    assert Backups.get_snapshot_by_id(snap2.id).id == snap2.id


@pytest.fixture(params=["instant_server_stop", "delayed_server_stop"])
def simulated_service_stopping_delay(request) -> float:
    if request.param == "instant_server_stop":
        return 0.0
    else:
        return 0.3


def test_backup_service_task(backups, dummy_service, simulated_service_stopping_delay):
    dummy_service.set_delay(simulated_service_stopping_delay)

    handle = start_backup(dummy_service.get_id())
    handle(blocking=True)

    snaps = Backups.get_snapshots(dummy_service)
    assert len(snaps) == 1

    id = dummy_service.get_id()
    job_type_id = f"services.{id}.backup"
    assert_job_finished(job_type_id, count=1)
    assert_job_has_run(job_type_id)
    assert_job_had_progress(job_type_id)


def test_forget_snapshot(backups, dummy_service):
    snap1 = Backups.back_up(dummy_service)
    snap2 = Backups.back_up(dummy_service)
    assert len(Backups.get_snapshots(dummy_service)) == 2

    Backups.forget_snapshot(snap2)
    assert len(Backups.get_snapshots(dummy_service)) == 1
    Backups.force_snapshot_cache_reload()
    assert len(Backups.get_snapshots(dummy_service)) == 1

    assert Backups.get_snapshots(dummy_service)[0].id == snap1.id

    Backups.forget_snapshot(snap1)
    assert len(Backups.get_snapshots(dummy_service)) == 0


def test_forget_nonexistent_snapshot(backups, dummy_service):
    bogus = Snapshot(
        id="gibberjibber",
        service_name="nohoho",
        created_at=datetime.now(timezone.utc),
        reason=BackupReason.EXPLICIT,
    )
    with pytest.raises(ValueError):
        Backups.forget_snapshot(bogus)


def test_backup_larger_file(backups, dummy_service):
    dir = path.join(dummy_service.get_folders()[0], "LARGEFILE")
    mega = 2**20
    make_large_file(dir, 100 * mega)

    handle = start_backup(dummy_service.get_id())
    handle(blocking=True)

    # results will be slightly different on different machines. if someone has troubles with it on their machine, consider dropping this test.
    id = dummy_service.get_id()
    job_type_id = f"services.{id}.backup"
    assert_job_finished(job_type_id, count=1)
    assert_job_has_run(job_type_id)
    updates = job_progress_updates(job_type_id)
    assert len(updates) > 3
    assert updates[int((len(updates) - 1) / 2.0)] > 10
    # clean up a bit
    remove(dir)


@pytest.fixture(params=["verify", "inplace"])
def restore_strategy(request) -> RestoreStrategy:
    if request.param == "verify":
        return RestoreStrategy.DOWNLOAD_VERIFY_OVERWRITE
    else:
        return RestoreStrategy.INPLACE


@pytest.fixture(params=["failed", "healthy", "fail_to_stop"])
def failed(request) -> str:
    return request.param


def test_restore_snapshot_task(
    backups, dummy_service, restore_strategy, simulated_service_stopping_delay, failed
):
    dummy_service.set_delay(simulated_service_stopping_delay)
    if failed == "failed":
        dummy_service.set_status(ServiceStatus.FAILED)

    if failed == "fail_to_stop":
        dummy_service.simulate_fail_to_stop(True)

    Backups.back_up(dummy_service)
    snaps = Backups.get_snapshots(dummy_service)
    assert len(snaps) == 1

    paths_to_nuke = service_files(dummy_service)
    contents = []

    for service_file in paths_to_nuke:
        with open(service_file, "r") as file:
            contents.append(file.read())

    for p in paths_to_nuke:
        remove(p)

    handle = restore_snapshot(snaps[0], restore_strategy)
    handle(blocking=True)

    for p, content in zip(paths_to_nuke, contents):
        assert path.exists(p)
        with open(p, "r") as file:
            assert file.read() == content

    snaps = Backups.get_snapshots(dummy_service)
    if restore_strategy == RestoreStrategy.INPLACE:
        assert len(snaps) == 2
        reasons = [snap.reason for snap in snaps]
        assert BackupReason.PRE_RESTORE in reasons
    else:
        assert len(snaps) == 1


def test_backup_unbackuppable(backups, dummy_service):
    dummy_service.set_backuppable(False)
    assert dummy_service.can_be_backed_up() is False
    with pytest.raises(ValueError):
        Backups.back_up(dummy_service)


# Storage
def test_snapshots_caching(backups, dummy_service):
    Backups.back_up(dummy_service)

    # we test indirectly that we do redis calls instead of shell calls
    start = datetime.now()
    for i in range(10):
        snapshots = Backups.get_snapshots(dummy_service)
        assert len(snapshots) == 1
    assert datetime.now() - start < timedelta(seconds=0.5)

    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1

    snap_to_uncache = cached_snapshots[0]
    Storage.delete_cached_snapshot(snap_to_uncache)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 0

    # We do not assume that no snapshots means we need to reload the cache
    snapshots = Backups.get_snapshots(dummy_service)
    assert len(snapshots) == 0
    # No cache reload happened
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 0


# Storage
def test_snapshot_cache_autoreloads(backups, dummy_service):
    Backups.back_up(dummy_service)

    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1
    snap_to_uncache = cached_snapshots[0]

    Storage.delete_cached_snapshot(snap_to_uncache)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 0

    # When we create a snapshot we do reload cache
    Backups.back_up(dummy_service)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 2
    assert snap_to_uncache in cached_snapshots

    Storage.delete_cached_snapshot(snap_to_uncache)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1

    # When we try to delete a snapshot we cannot find in cache, it is ok and we do reload cache
    Backups.forget_snapshot(snap_to_uncache)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1
    assert snap_to_uncache not in cached_snapshots


def lowlevel_forget(snapshot_id):
    Backups.provider().backupper.forget_snapshot(snapshot_id)


# Storage
def test_snapshots_cache_invalidation(backups, dummy_service):
    Backups.back_up(dummy_service)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1

    Storage.invalidate_snapshot_storage()
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 0

    Backups.force_snapshot_cache_reload()
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1
    snap = cached_snapshots[0]

    lowlevel_forget(snap.id)
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 1

    Backups.force_snapshot_cache_reload()
    cached_snapshots = Storage.get_cached_snapshots()
    assert len(cached_snapshots) == 0


# Storage
def test_init_tracking_caching(backups, raw_dummy_service):
    assert Storage.has_init_mark() is True
    Backups.reset()
    assert Storage.has_init_mark() is False

    Storage.mark_as_init()

    assert Storage.has_init_mark() is True
    assert Backups.is_initted() is True


# Storage
def test_init_tracking_caching2(backups, tmpdir):
    assert Storage.has_init_mark() is True
    Backups.reset()
    assert Storage.has_init_mark() is False
    separate_dir = tmpdir / "out_of_the_way"
    prepare_localfile_backups(separate_dir)
    assert Storage.has_init_mark() is False

    Backups.init_repo()

    assert Storage.has_init_mark() is True


# Storage
def test_provider_storage(backups):
    test_login = "ID"
    test_key = "KEY"
    test_location = "selprivacy_bin"

    old_provider = Backups.provider()
    assert old_provider is not None

    assert not isinstance(old_provider, Backblaze)
    assert old_provider.login != test_login
    assert old_provider.key != test_key
    assert old_provider.location != test_location

    test_provider = Backups._construct_provider(
        kind=BackupProvider.BACKBLAZE, login="ID", key=test_key, location=test_location
    )

    assert isinstance(test_provider, Backblaze)
    assert get_kind(test_provider) == "BACKBLAZE"
    assert test_provider.login == test_login
    assert test_provider.key == test_key
    assert test_provider.location == test_location

    Storage.store_provider(test_provider)

    restored_provider_model = Storage.load_provider()
    assert restored_provider_model.kind == "BACKBLAZE"
    assert restored_provider_model.login == test_login
    assert restored_provider_model.key == test_key
    assert restored_provider_model.location == test_location

    restored_provider = Backups._load_provider_redis()
    assert isinstance(restored_provider, Backblaze)
    assert restored_provider.login == test_login
    assert restored_provider.key == test_key
    assert restored_provider.location == test_location

    # Revert our mess so we can teardown ok
    Storage.store_provider(old_provider)


def test_sync(dummy_service):
    src = dummy_service.get_folders()[0]
    dst = dummy_service.get_folders()[1]
    old_files_src = set(listdir(src))
    old_files_dst = set(listdir(dst))
    assert old_files_src != old_files_dst

    sync(src, dst)
    new_files_src = set(listdir(src))
    new_files_dst = set(listdir(dst))
    assert new_files_src == old_files_src
    assert new_files_dst == new_files_src


def test_sync_nonexistent_src(dummy_service):
    src = "/var/lib/nonexistentFluffyBunniesOfUnix"
    dst = dummy_service.get_folders()[1]

    with pytest.raises(ValueError):
        sync(src, dst)


def test_move_blocks_backups(backups, dummy_service, restore_strategy):
    snap = Backups.back_up(dummy_service)
    job = Jobs.add(
        type_id=f"services.{dummy_service.get_id()}.move",
        name="Move Dummy",
        description=f"Moving Dummy data to the Rainbow Land",
        status=JobStatus.RUNNING,
    )

    with pytest.raises(ValueError):
        Backups.back_up(dummy_service)

    with pytest.raises(ValueError):
        Backups.restore_snapshot(snap, restore_strategy)


def test_double_lock_unlock(backups, dummy_service):
    # notice that introducing stale locks is only safe for other tests if we erase repo in between
    # which we do at the time of writing this test

    Backups.provider().backupper.lock()
    with pytest.raises(ValueError):
        Backups.provider().backupper.lock()

    Backups.provider().backupper.unlock()
    Backups.provider().backupper.lock()

    Backups.provider().backupper.unlock()
    Backups.provider().backupper.unlock()


def test_operations_while_locked(backups, dummy_service):
    # Stale lock prevention test

    # consider making it fully at the level of backupper?
    # because this is where prevention lives?
    # Backups singleton is here only so that we can run this against B2, S3 and whatever
    # But maybe it is not necessary (if restic treats them uniformly enough)

    Backups.provider().backupper.lock()
    snap = Backups.back_up(dummy_service)
    assert snap is not None

    Backups.provider().backupper.lock()
    # using lowlevel to make sure no caching interferes
    assert Backups.provider().backupper.is_initted() is True

    Backups.provider().backupper.lock()
    assert Backups.snapshot_restored_size(snap.id) > 0

    Backups.provider().backupper.lock()
    Backups.restore_snapshot(snap)

    Backups.provider().backupper.lock()
    Backups.forget_snapshot(snap)

    Backups.provider().backupper.lock()
    assert Backups.provider().backupper.get_snapshots() == []

    # check that no locks were left
    Backups.provider().backupper.lock()
    Backups.provider().backupper.unlock()


# a paranoid check to weed out problems with tempdirs that are not dependent on us
def test_tempfile():
    with tempfile.TemporaryDirectory() as temp:
        assert path.exists(temp)
    assert not path.exists(temp)


# Storage
def test_cache_invalidaton_task(backups, dummy_service):
    Backups.back_up(dummy_service)
    assert len(Storage.get_cached_snapshots()) == 1

    # Does not trigger resync
    Storage.invalidate_snapshot_storage()
    assert Storage.get_cached_snapshots() == []

    reload_snapshot_cache()
    assert len(Storage.get_cached_snapshots()) == 1


def test_service_manager_backup_snapshot_persists(backups, generic_userdata, dkim_file):
    # There was a bug with snapshot disappearance due to post_restore hooks, checking for that
    manager = ServiceManager.get_service_by_id("api")
    assert manager is not None

    snapshot = Backups.back_up(manager)

    Backups.force_snapshot_cache_reload()
    ids = [snap.id for snap in Backups.get_all_snapshots()]
    assert snapshot.id in ids


def test_service_manager_backs_up_without_crashing(
    backups, generic_userdata, dkim_file, dummy_service
):
    """
    Service manager is special and needs testing.
    """
    manager = ServiceManager.get_service_by_id("api")
    assert manager is not None

    snapshot = Backups.back_up(manager)
    Backups.restore_snapshot(snapshot)