google/cloud/spanner_v1/snapshot.py

# Copyright 2016 Google LLC All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Model a set of read-only queries to a database as a snapshot."""

from datetime import datetime
import functools
import threading
from google.protobuf.struct_pb2 import Struct
from google.cloud.spanner_v1 import ExecuteSqlRequest
from google.cloud.spanner_v1 import ReadRequest
from google.cloud.spanner_v1 import TransactionOptions
from google.cloud.spanner_v1 import TransactionSelector
from google.cloud.spanner_v1 import PartitionOptions
from google.cloud.spanner_v1 import PartitionQueryRequest
from google.cloud.spanner_v1 import PartitionReadRequest

from google.api_core.exceptions import InternalServerError
from google.api_core.exceptions import ServiceUnavailable
from google.api_core.exceptions import InvalidArgument
from google.api_core import gapic_v1
from google.cloud.spanner_v1._helpers import (
    _make_value_pb,
    _merge_query_options,
    _metadata_with_prefix,
    _metadata_with_leader_aware_routing,
    _retry,
    _check_rst_stream_error,
    _SessionWrapper,
)
from google.cloud.spanner_v1._opentelemetry_tracing import trace_call
from google.cloud.spanner_v1.streamed import StreamedResultSet
from google.cloud.spanner_v1 import RequestOptions

_STREAM_RESUMPTION_INTERNAL_ERROR_MESSAGES = (
    "RST_STREAM",
    "Received unexpected EOS on DATA frame from server",
)


def _restart_on_unavailable(
    method,
    request,
    trace_name=None,
    session=None,
    attributes=None,
    transaction=None,
    transaction_selector=None,
    observability_options=None,
):
    """Restart iteration after :exc:`.ServiceUnavailable`.

    :type method: callable
    :param method: function returning iterator

    :type request: proto
    :param request: request proto to call the method with

    :type transaction: :class:`google.cloud.spanner_v1.snapshot._SnapshotBase`
    :param transaction: Snapshot or Transaction class object based on the type of transaction

    :type transaction_selector: :class:`transaction_pb2.TransactionSelector`
    :param transaction_selector: Transaction selector object to be used in request if transaction is not passed,
    if both transaction_selector and transaction are passed, then transaction is given priority.
    """

    resume_token = b""
    item_buffer = []

    if transaction is not None:
        transaction_selector = transaction._make_txn_selector()
    elif transaction_selector is None:
        raise InvalidArgument(
            "Either transaction or transaction_selector should be set"
        )

    request.transaction = transaction_selector
    iterator = None

    while True:
        try:
            if iterator is None:
                with trace_call(
                    trace_name,
                    session,
                    attributes,
                    observability_options=observability_options,
                ):
                    iterator = method(request=request)
            for item in iterator:
                item_buffer.append(item)
                # Setting the transaction id because the transaction begin was inlined for first rpc.
                if (
                    transaction is not None
                    and transaction._transaction_id is None
                    and item.metadata is not None
                    and item.metadata.transaction is not None
                    and item.metadata.transaction.id is not None
                ):
                    transaction._transaction_id = item.metadata.transaction.id
                if item.resume_token:
                    resume_token = item.resume_token
                    break
        except ServiceUnavailable:
            del item_buffer[:]
            with trace_call(
                trace_name,
                session,
                attributes,
                observability_options=observability_options,
            ):
                request.resume_token = resume_token
                if transaction is not None:
                    transaction_selector = transaction._make_txn_selector()
                request.transaction = transaction_selector
                iterator = method(request=request)
            continue
        except InternalServerError as exc:
            resumable_error = any(
                resumable_message in exc.message
                for resumable_message in _STREAM_RESUMPTION_INTERNAL_ERROR_MESSAGES
            )
            if not resumable_error:
                raise
            del item_buffer[:]
            with trace_call(
                trace_name,
                session,
                attributes,
                observability_options=observability_options,
            ):
                request.resume_token = resume_token
                if transaction is not None:
                    transaction_selector = transaction._make_txn_selector()
                request.transaction = transaction_selector
                iterator = method(request=request)
            continue

        if len(item_buffer) == 0:
            break

        for item in item_buffer:
            yield item

        del item_buffer[:]


class _SnapshotBase(_SessionWrapper):
    """Base class for Snapshot.

    Allows reuse of API request methods with different transaction selector.

    :type session: :class:`~google.cloud.spanner_v1.session.Session`
    :param session: the session used to perform the commit
    """

    _multi_use = False
    _read_only: bool = True
    _transaction_id = None
    _read_request_count = 0
    _execute_sql_count = 0
    _lock = threading.Lock()

    def _make_txn_selector(self):
        """Helper for :meth:`read` / :meth:`execute_sql`.

        Subclasses must override, returning an instance of
        :class:`transaction_pb2.TransactionSelector`
        appropriate for making ``read`` / ``execute_sql`` requests

        :raises: NotImplementedError, always
        """
        raise NotImplementedError

    def read(
        self,
        table,
        columns,
        keyset,
        index="",
        limit=0,
        partition=None,
        request_options=None,
        data_boost_enabled=False,
        directed_read_options=None,
        *,
        retry=gapic_v1.method.DEFAULT,
        timeout=gapic_v1.method.DEFAULT,
        column_info=None,
        lazy_decode=False,
    ):
        """Perform a ``StreamingRead`` API request for rows in a table.

        :type table: str
        :param table: name of the table from which to fetch data

        :type columns: list of str
        :param columns: names of columns to be retrieved

        :type keyset: :class:`~google.cloud.spanner_v1.keyset.KeySet`
        :param keyset: keys / ranges identifying rows to be retrieved

        :type index: str
        :param index: (Optional) name of index to use, rather than the
                      table's primary key

        :type limit: int
        :param limit: (Optional) maximum number of rows to return.
                      Incompatible with ``partition``.

        :type partition: bytes
        :param partition: (Optional) one of the partition tokens returned
                          from :meth:`partition_read`.  Incompatible with
                          ``limit``.

        :type request_options:
            :class:`google.cloud.spanner_v1.types.RequestOptions`
        :param request_options:
                (Optional) Common options for this request.
                If a dict is provided, it must be of the same form as the protobuf
                message :class:`~google.cloud.spanner_v1.types.RequestOptions`.
                Please note, the `transactionTag` setting will be ignored for
                snapshot as it's not supported for read-only transactions.

        :type retry: :class:`~google.api_core.retry.Retry`
        :param retry: (Optional) The retry settings for this request.

        :type timeout: float
        :param timeout: (Optional) The timeout for this request.

        :type data_boost_enabled:
        :param data_boost_enabled:
                (Optional) If this is for a partitioned read and this field is
                set ``true``, the request will be executed via offline access.
                If the field is set to ``true`` but the request does not set
                ``partition_token``, the API will return an
                ``INVALID_ARGUMENT`` error.

        :type directed_read_options: :class:`~google.cloud.spanner_v1.DirectedReadOptions`
            or :class:`dict`
        :param directed_read_options: (Optional) Request level option used to set the directed_read_options
            for all ReadRequests and ExecuteSqlRequests that indicates which replicas
            or regions should be used for non-transactional reads or queries.

        :type column_info: dict
        :param column_info: (Optional) dict of mapping between column names and additional column information.
            An object where column names as keys and custom objects as corresponding
            values for deserialization. It's specifically useful for data types like
            protobuf where deserialization logic is on user-specific code. When provided,
            the custom object enables deserialization of backend-received column data.
            If not provided, data remains serialized as bytes for Proto Messages and
            integer for Proto Enums.

        :type lazy_decode: bool
        :param lazy_decode:
            (Optional) If this argument is set to ``true``, the iterator
            returns the underlying protobuf values instead of decoded Python
            objects. This reduces the time that is needed to iterate through
            large result sets. The application is responsible for decoding
            the data that is needed. The returned row iterator contains two
            functions that can be used for this. ``iterator.decode_row(row)``
            decodes all the columns in the given row to an array of Python
            objects. ``iterator.decode_column(row, column_index)`` decodes one
            specific column in the given row.

        :rtype: :class:`~google.cloud.spanner_v1.streamed.StreamedResultSet`
        :returns: a result set instance which can be used to consume rows.

        :raises ValueError:
            for reuse of single-use snapshots, or if a transaction ID is
            already pending for multiple-use snapshots.
        """
        if self._read_request_count > 0:
            if not self._multi_use:
                raise ValueError("Cannot re-use single-use snapshot.")
            if self._transaction_id is None and self._read_only:
                raise ValueError("Transaction ID pending.")

        database = self._session._database
        api = database.spanner_api
        metadata = _metadata_with_prefix(database.name)
        if not self._read_only and database._route_to_leader_enabled:
            metadata.append(
                _metadata_with_leader_aware_routing(database._route_to_leader_enabled)
            )

        if request_options is None:
            request_options = RequestOptions()
        elif type(request_options) is dict:
            request_options = RequestOptions(request_options)

        if self._read_only:
            # Transaction tags are not supported for read only transactions.
            request_options.transaction_tag = None
            if (
                directed_read_options is None
                and database._directed_read_options is not None
            ):
                directed_read_options = database._directed_read_options
        elif self.transaction_tag is not None:
            request_options.transaction_tag = self.transaction_tag

        request = ReadRequest(
            session=self._session.name,
            table=table,
            columns=columns,
            key_set=keyset._to_pb(),
            index=index,
            limit=limit,
            partition_token=partition,
            request_options=request_options,
            data_boost_enabled=data_boost_enabled,
            directed_read_options=directed_read_options,
        )
        restart = functools.partial(
            api.streaming_read,
            request=request,
            metadata=metadata,
            retry=retry,
            timeout=timeout,
        )

        trace_attributes = {"table_id": table, "columns": columns}
        observability_options = getattr(database, "observability_options", None)

        if self._transaction_id is None:
            # lock is added to handle the inline begin for first rpc
            with self._lock:
                iterator = _restart_on_unavailable(
                    restart,
                    request,
                    f"CloudSpanner.{type(self).__name__}.read",
                    self._session,
                    trace_attributes,
                    transaction=self,
                    observability_options=observability_options,
                )
                self._read_request_count += 1
                if self._multi_use:
                    return StreamedResultSet(
                        iterator,
                        source=self,
                        column_info=column_info,
                        lazy_decode=lazy_decode,
                    )
                else:
                    return StreamedResultSet(
                        iterator, column_info=column_info, lazy_decode=lazy_decode
                    )
        else:
            iterator = _restart_on_unavailable(
                restart,
                request,
                f"CloudSpanner.{type(self).__name__}.read",
                self._session,
                trace_attributes,
                transaction=self,
                observability_options=observability_options,
            )

        self._read_request_count += 1
        self._session._last_use_time = datetime.now()

        if self._multi_use:
            return StreamedResultSet(
                iterator, source=self, column_info=column_info, lazy_decode=lazy_decode
            )
        else:
            return StreamedResultSet(
                iterator, column_info=column_info, lazy_decode=lazy_decode
            )

    def execute_sql(
        self,
        sql,
        params=None,
        param_types=None,
        query_mode=None,
        query_options=None,
        request_options=None,
        partition=None,
        retry=gapic_v1.method.DEFAULT,
        timeout=gapic_v1.method.DEFAULT,
        data_boost_enabled=False,
        directed_read_options=None,
        column_info=None,
        lazy_decode=False,
    ):
        """Perform an ``ExecuteStreamingSql`` API request.

        :type sql: str
        :param sql: SQL query statement

        :type params: dict, {str -> column value}
        :param params: values for parameter replacement.  Keys must match
                       the names used in ``sql``.

        :type param_types: dict[str -> Union[dict, .types.Type]]
        :param param_types:
            (Optional) maps explicit types for one or more param values;
            required if parameters are passed.

        :type query_mode:
            :class:`~google.cloud.spanner_v1.types.ExecuteSqlRequest.QueryMode`
        :param query_mode: Mode governing return of results / query plan.
            See:
            `QueryMode <https://cloud.google.com/spanner/reference/rpc/google.spanner.v1#google.spanner.v1.ExecuteSqlRequest.QueryMode>`_.

        :type query_options:
            :class:`~google.cloud.spanner_v1.types.ExecuteSqlRequest.QueryOptions`
                or :class:`dict`
        :param query_options:
                (Optional) Query optimizer configuration to use for the given query.
                If a dict is provided, it must be of the same form as the protobuf
                message :class:`~google.cloud.spanner_v1.types.QueryOptions`

        :type request_options:
            :class:`google.cloud.spanner_v1.types.RequestOptions`
        :param request_options:
                (Optional) Common options for this request.
                If a dict is provided, it must be of the same form as the protobuf
                message :class:`~google.cloud.spanner_v1.types.RequestOptions`.

        :type partition: bytes
        :param partition: (Optional) one of the partition tokens returned
                          from :meth:`partition_query`.

        :rtype: :class:`~google.cloud.spanner_v1.streamed.StreamedResultSet`
        :returns: a result set instance which can be used to consume rows.

        :type retry: :class:`~google.api_core.retry.Retry`
        :param retry: (Optional) The retry settings for this request.

        :type timeout: float
        :param timeout: (Optional) The timeout for this request.

        :type data_boost_enabled:
        :param data_boost_enabled:
                (Optional) If this is for a partitioned query and this field is
                set ``true``, the request will be executed via offline access.
                If the field is set to ``true`` but the request does not set
                ``partition_token``, the API will return an
                ``INVALID_ARGUMENT`` error.

        :type directed_read_options: :class:`~google.cloud.spanner_v1.DirectedReadOptions`
            or :class:`dict`
        :param directed_read_options: (Optional) Request level option used to set the directed_read_options
            for all ReadRequests and ExecuteSqlRequests that indicates which replicas
            or regions should be used for non-transactional reads or queries.

        :type column_info: dict
        :param column_info: (Optional) dict of mapping between column names and additional column information.
            An object where column names as keys and custom objects as corresponding
            values for deserialization. It's specifically useful for data types like
            protobuf where deserialization logic is on user-specific code. When provided,
            the custom object enables deserialization of backend-received column data.
            If not provided, data remains serialized as bytes for Proto Messages and
            integer for Proto Enums.

        :type lazy_decode: bool
        :param lazy_decode:
            (Optional) If this argument is set to ``true``, the iterator
            returns the underlying protobuf values instead of decoded Python
            objects. This reduces the time that is needed to iterate through
            large result sets. The application is responsible for decoding
            the data that is needed. The returned row iterator contains two
            functions that can be used for this. ``iterator.decode_row(row)``
            decodes all the columns in the given row to an array of Python
            objects. ``iterator.decode_column(row, column_index)`` decodes one
            specific column in the given row.

        :raises ValueError:
            for reuse of single-use snapshots, or if a transaction ID is
            already pending for multiple-use snapshots.
        """
        if self._read_request_count > 0:
            if not self._multi_use:
                raise ValueError("Cannot re-use single-use snapshot.")
            if self._transaction_id is None and self._read_only:
                raise ValueError("Transaction ID pending.")

        if params is not None:
            params_pb = Struct(
                fields={key: _make_value_pb(value) for key, value in params.items()}
            )
        else:
            params_pb = {}

        database = self._session._database
        metadata = _metadata_with_prefix(database.name)
        if not self._read_only and database._route_to_leader_enabled:
            metadata.append(
                _metadata_with_leader_aware_routing(database._route_to_leader_enabled)
            )

        api = database.spanner_api

        # Query-level options have higher precedence than client-level and
        # environment-level options
        default_query_options = database._instance._client._query_options
        query_options = _merge_query_options(default_query_options, query_options)

        if request_options is None:
            request_options = RequestOptions()
        elif type(request_options) is dict:
            request_options = RequestOptions(request_options)
        if self._read_only:
            # Transaction tags are not supported for read only transactions.
            request_options.transaction_tag = None
            if (
                directed_read_options is None
                and database._directed_read_options is not None
            ):
                directed_read_options = database._directed_read_options
        elif self.transaction_tag is not None:
            request_options.transaction_tag = self.transaction_tag

        request = ExecuteSqlRequest(
            session=self._session.name,
            sql=sql,
            params=params_pb,
            param_types=param_types,
            query_mode=query_mode,
            partition_token=partition,
            seqno=self._execute_sql_count,
            query_options=query_options,
            request_options=request_options,
            data_boost_enabled=data_boost_enabled,
            directed_read_options=directed_read_options,
        )
        restart = functools.partial(
            api.execute_streaming_sql,
            request=request,
            metadata=metadata,
            retry=retry,
            timeout=timeout,
        )

        trace_attributes = {"db.statement": sql}
        observability_options = getattr(database, "observability_options", None)

        if self._transaction_id is None:
            # lock is added to handle the inline begin for first rpc
            with self._lock:
                return self._get_streamed_result_set(
                    restart,
                    request,
                    trace_attributes,
                    column_info,
                    observability_options,
                    lazy_decode=lazy_decode,
                )
        else:
            return self._get_streamed_result_set(
                restart,
                request,
                trace_attributes,
                column_info,
                observability_options,
                lazy_decode=lazy_decode,
            )

    def _get_streamed_result_set(
        self,
        restart,
        request,
        trace_attributes,
        column_info,
        observability_options=None,
        lazy_decode=False,
    ):
        iterator = _restart_on_unavailable(
            restart,
            request,
            f"CloudSpanner.{type(self).__name__}.execute_sql",
            self._session,
            trace_attributes,
            transaction=self,
            observability_options=observability_options,
        )
        self._read_request_count += 1
        self._execute_sql_count += 1

        if self._multi_use:
            return StreamedResultSet(
                iterator, source=self, column_info=column_info, lazy_decode=lazy_decode
            )
        else:
            return StreamedResultSet(
                iterator, column_info=column_info, lazy_decode=lazy_decode
            )

    def partition_read(
        self,
        table,
        columns,
        keyset,
        index="",
        partition_size_bytes=None,
        max_partitions=None,
        *,
        retry=gapic_v1.method.DEFAULT,
        timeout=gapic_v1.method.DEFAULT,
    ):
        """Perform a ``PartitionRead`` API request for rows in a table.

        :type table: str
        :param table: name of the table from which to fetch data

        :type columns: list of str
        :param columns: names of columns to be retrieved

        :type keyset: :class:`~google.cloud.spanner_v1.keyset.KeySet`
        :param keyset: keys / ranges identifying rows to be retrieved

        :type index: str
        :param index: (Optional) name of index to use, rather than the
                      table's primary key

        :type partition_size_bytes: int
        :param partition_size_bytes:
            (Optional) desired size for each partition generated.  The service
            uses this as a hint, the actual partition size may differ.

        :type max_partitions: int
        :param max_partitions:
            (Optional) desired maximum number of partitions generated. The
            service uses this as a hint, the actual number of partitions may
            differ.

        :type retry: :class:`~google.api_core.retry.Retry`
        :param retry: (Optional) The retry settings for this request.

        :type timeout: float
        :param timeout: (Optional) The timeout for this request.

        :rtype: iterable of bytes
        :returns: a sequence of partition tokens

        :raises ValueError:
            for single-use snapshots, or if a transaction ID is
            already associated with the snapshot.
        """
        if not self._multi_use:
            raise ValueError("Cannot use single-use snapshot.")

        if self._transaction_id is None:
            raise ValueError("Transaction not started.")

        database = self._session._database
        api = database.spanner_api
        metadata = _metadata_with_prefix(database.name)
        if database._route_to_leader_enabled:
            metadata.append(
                _metadata_with_leader_aware_routing(database._route_to_leader_enabled)
            )
        transaction = self._make_txn_selector()
        partition_options = PartitionOptions(
            partition_size_bytes=partition_size_bytes, max_partitions=max_partitions
        )
        request = PartitionReadRequest(
            session=self._session.name,
            table=table,
            columns=columns,
            key_set=keyset._to_pb(),
            transaction=transaction,
            index=index,
            partition_options=partition_options,
        )

        trace_attributes = {"table_id": table, "columns": columns}
        can_include_index = (index != "") and (index is not None)
        if can_include_index:
            trace_attributes["index"] = index

        with trace_call(
            f"CloudSpanner.{type(self).__name__}.partition_read",
            self._session,
            extra_attributes=trace_attributes,
            observability_options=getattr(database, "observability_options", None),
        ):
            method = functools.partial(
                api.partition_read,
                request=request,
                metadata=metadata,
                retry=retry,
                timeout=timeout,
            )
            response = _retry(
                method,
                allowed_exceptions={InternalServerError: _check_rst_stream_error},
            )

        return [partition.partition_token for partition in response.partitions]

    def partition_query(
        self,
        sql,
        params=None,
        param_types=None,
        partition_size_bytes=None,
        max_partitions=None,
        *,
        retry=gapic_v1.method.DEFAULT,
        timeout=gapic_v1.method.DEFAULT,
    ):
        """Perform a ``PartitionQuery`` API request.

        :type sql: str
        :param sql: SQL query statement

        :type params: dict, {str -> column value}
        :param params: values for parameter replacement.  Keys must match
                       the names used in ``sql``.

        :type param_types: dict[str -> Union[dict, .types.Type]]
        :param param_types:
            (Optional) maps explicit types for one or more param values;
            required if parameters are passed.

        :type partition_size_bytes: int
        :param partition_size_bytes:
            (Optional) desired size for each partition generated.  The service
            uses this as a hint, the actual partition size may differ.

        :type max_partitions: int
        :param max_partitions:
            (Optional) desired maximum number of partitions generated. The
            service uses this as a hint, the actual number of partitions may
            differ.

        :type retry: :class:`~google.api_core.retry.Retry`
        :param retry: (Optional) The retry settings for this request.

        :type timeout: float
        :param timeout: (Optional) The timeout for this request.

        :rtype: iterable of bytes
        :returns: a sequence of partition tokens

        :raises ValueError:
            for single-use snapshots, or if a transaction ID is
            already associated with the snapshot.
        """
        if not self._multi_use:
            raise ValueError("Cannot use single-use snapshot.")

        if self._transaction_id is None:
            raise ValueError("Transaction not started.")

        if params is not None:
            params_pb = Struct(
                fields={key: _make_value_pb(value) for (key, value) in params.items()}
            )
        else:
            params_pb = Struct()

        database = self._session._database
        api = database.spanner_api
        metadata = _metadata_with_prefix(database.name)
        if database._route_to_leader_enabled:
            metadata.append(
                _metadata_with_leader_aware_routing(database._route_to_leader_enabled)
            )
        transaction = self._make_txn_selector()
        partition_options = PartitionOptions(
            partition_size_bytes=partition_size_bytes, max_partitions=max_partitions
        )
        request = PartitionQueryRequest(
            session=self._session.name,
            sql=sql,
            transaction=transaction,
            params=params_pb,
            param_types=param_types,
            partition_options=partition_options,
        )

        trace_attributes = {"db.statement": sql}
        with trace_call(
            f"CloudSpanner.{type(self).__name__}.partition_query",
            self._session,
            trace_attributes,
            observability_options=getattr(database, "observability_options", None),
        ):
            method = functools.partial(
                api.partition_query,
                request=request,
                metadata=metadata,
                retry=retry,
                timeout=timeout,
            )
            response = _retry(
                method,
                allowed_exceptions={InternalServerError: _check_rst_stream_error},
            )

        return [partition.partition_token for partition in response.partitions]


class Snapshot(_SnapshotBase):
    """Allow a set of reads / SQL statements with shared staleness.

    See
    https://cloud.google.com/spanner/reference/rpc/google.spanner.v1#google.spanner.v1.TransactionOptions.ReadOnly

    If no options are passed, reads will use the ``strong`` model, reading
    at a timestamp where all previously committed transactions are visible.

    :type session: :class:`~google.cloud.spanner_v1.session.Session`
    :param session: The session used to perform the commit.

    :type read_timestamp: :class:`datetime.datetime`
    :param read_timestamp: Execute all reads at the given timestamp.

    :type min_read_timestamp: :class:`datetime.datetime`
    :param min_read_timestamp: Execute all reads at a
                               timestamp >= ``min_read_timestamp``.

    :type max_staleness: :class:`datetime.timedelta`
    :param max_staleness: Read data at a
                          timestamp >= NOW - ``max_staleness`` seconds.

    :type exact_staleness: :class:`datetime.timedelta`
    :param exact_staleness: Execute all reads at a timestamp that is
                            ``exact_staleness`` old.

    :type multi_use: :class:`bool`
    :param multi_use: If true, multiple :meth:`read` / :meth:`execute_sql`
                      calls can be performed with the snapshot in the
                      context of a read-only transaction, used to ensure
                      isolation / consistency. Incompatible with
                      ``max_staleness`` and ``min_read_timestamp``.
    """

    def __init__(
        self,
        session,
        read_timestamp=None,
        min_read_timestamp=None,
        max_staleness=None,
        exact_staleness=None,
        multi_use=False,
        transaction_id=None,
    ):
        super(Snapshot, self).__init__(session)
        opts = [read_timestamp, min_read_timestamp, max_staleness, exact_staleness]
        flagged = [opt for opt in opts if opt is not None]

        if len(flagged) > 1:
            raise ValueError("Supply zero or one options.")

        if multi_use:
            if min_read_timestamp is not None or max_staleness is not None:
                raise ValueError(
                    "'multi_use' is incompatible with "
                    "'min_read_timestamp' / 'max_staleness'"
                )

        self._transaction_read_timestamp = None
        self._strong = len(flagged) == 0
        self._read_timestamp = read_timestamp
        self._min_read_timestamp = min_read_timestamp
        self._max_staleness = max_staleness
        self._exact_staleness = exact_staleness
        self._multi_use = multi_use
        self._transaction_id = transaction_id

    def _make_txn_selector(self):
        """Helper for :meth:`read`."""
        if self._transaction_id is not None:
            return TransactionSelector(id=self._transaction_id)

        if self._read_timestamp:
            key = "read_timestamp"
            value = self._read_timestamp
        elif self._min_read_timestamp:
            key = "min_read_timestamp"
            value = self._min_read_timestamp
        elif self._max_staleness:
            key = "max_staleness"
            value = self._max_staleness
        elif self._exact_staleness:
            key = "exact_staleness"
            value = self._exact_staleness
        else:
            key = "strong"
            value = True

        options = TransactionOptions(
            read_only=TransactionOptions.ReadOnly(
                **{key: value, "return_read_timestamp": True}
            )
        )

        if self._multi_use:
            return TransactionSelector(begin=options)
        else:
            return TransactionSelector(single_use=options)

    def begin(self):
        """Begin a read-only transaction on the database.

        :rtype: bytes
        :returns: the ID for the newly-begun transaction.

        :raises ValueError:
            if the transaction is already begun, committed, or rolled back.
        """
        if not self._multi_use:
            raise ValueError("Cannot call 'begin' on single-use snapshots")

        if self._transaction_id is not None:
            raise ValueError("Read-only transaction already begun")

        if self._read_request_count > 0:
            raise ValueError("Read-only transaction already pending")

        database = self._session._database
        api = database.spanner_api
        metadata = _metadata_with_prefix(database.name)
        if not self._read_only and database._route_to_leader_enabled:
            metadata.append(
                (_metadata_with_leader_aware_routing(database._route_to_leader_enabled))
            )
        txn_selector = self._make_txn_selector()
        with trace_call(
            f"CloudSpanner.{type(self).__name__}.begin",
            self._session,
            observability_options=getattr(database, "observability_options", None),
        ):
            method = functools.partial(
                api.begin_transaction,
                session=self._session.name,
                options=txn_selector.begin,
                metadata=metadata,
            )
            response = _retry(
                method,
                allowed_exceptions={InternalServerError: _check_rst_stream_error},
            )
        self._transaction_id = response.id
        self._transaction_read_timestamp = response.read_timestamp
        return self._transaction_id