toil 8.0.0__py3-none-any.whl → 8.1.0b1__py3-none-any.whl

toil/lib/history.py

@@ -0,0 +1,1271 @@
+# Copyright (C) 2024 Regents of the University of California
+#
+# 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.
+
+"""
+Contains tools for tracking history.
+"""
+
+from contextlib import contextmanager
+import logging
+import os
+import sqlite3
+import sys
+import threading
+import time
+import uuid
+from dataclasses import dataclass
+from typing import Any, Iterable, Iterator, Optional, TypeVar, Callable
+
+from toil.lib.io import get_toil_home
+from toil.lib.retry import ErrorCondition, retry
+
+logger = logging.getLogger(__name__)
+
+class HistoryDatabaseSchemaTooNewError(RuntimeError):
+    """
+    Raised when we would write to the history database, but its schema is too
+    new for us to understand.
+    """
+    pass
+
+@dataclass
+class WorkflowSummary:
+    """
+    Data class holding summary information for a workflow.
+
+    Represents all the attempts to execute one run of a workflow.
+    """
+    id: str
+    name: Optional[str]
+    job_store: str
+    total_attempts: int
+    total_job_attempts: int
+    succeeded: bool
+    start_time: Optional[float]
+    """
+    Time when the first workflow attempt started, in seconds since epoch.
+
+    None if there are no attempts recorded.
+    """
+    runtime: Optional[float]
+    """
+    Time from the first workflow attempt's start to the last one's end, in seconds.
+
+    None if there are no attempts recorded.
+    """
+    trs_spec: Optional[str]
+
+@dataclass
+class WorkflowAttemptSummary:
+    """
+    Data class holding summary information for a workflow attempt.
+
+    Helpfully includes the workflow metadata for Dockstore.
+    """
+    workflow_id: str
+    attempt_number: int
+    succeeded: bool
+    start_time: float
+    runtime: float
+    submitted_to_dockstore: bool
+    batch_system: Optional[str]
+    caching: Optional[bool]
+    toil_version: Optional[str]
+    python_version: Optional[str]
+    platform_system: Optional[str]
+    platform_machine: Optional[str]
+    workflow_job_store: str
+    workflow_trs_spec: Optional[str]
+
+@dataclass
+class JobAttemptSummary:
+    """
+    Data class holding summary information for a job attempt within a known
+    workflow attempt.
+    """
+    id: str
+    job_name: str
+    succeeded: bool
+    start_time: float
+    runtime: float
+    submitted_to_dockstore: bool
+    cores: Optional[float]
+    cpu_seconds: Optional[float]
+    memory_bytes: Optional[int]
+    disk_bytes: Optional[int]
+
+
+RT = TypeVar("RT")
+
+def db_retry(function: Callable[..., RT]) -> Callable[..., RT]:
+    """
+    Decorate a function with the appropriate retries for accessing the database.
+    """
+    return retry(
+        infinite_retries=True,
+        errors=[
+            ErrorCondition(
+                error=sqlite3.OperationalError, error_message_must_include="is locked"
+            )
+        ],
+    )(function)
+
+class HistoryManager:
+    """
+    Class responsible for managing the history of Toil runs.
+    """
+
+    # Should workflow run history be recorded?
+    WORKFLOW_HISTORY_ENABLED = True
+    # Should job history be recorded? Can only be true if
+    # WORKFLOW_HISTORY_ENABLED is also true.
+    #
+    # TODO: When Dockstore can take job metrics alongside whole-workflow
+    # metrics, and we've tested to make sure history recording doesn't slow
+    # down our leader job processing rate, turn on actual job history logging.
+    JOB_HISTORY_ENABLED = False
+
+    # For testing, we can move the database path for the class.
+    database_path_override: Optional[str] = None
+
+    @classmethod
+    def database_path(cls) -> str:
+        """
+        Get the path at which the database we store history in lives.
+        """
+        if cls.database_path_override is not None:
+            # Under test, we can use a temporary path.
+            return cls.database_path_override
+
+        return os.path.join(get_toil_home(), "history.sqlite")
+
+    @classmethod
+    def connection(cls) -> sqlite3.Connection:
+        """
+        Connect to the history database.
+
+        Caller must not actually use the connection without using
+        ensure_tables() to protect reads and updates.
+
+        Must be called from inside a top-level method marked @db_retry.
+
+        The connection will be in DEFERRED isolation_level, with autocommit off
+        on Python versions that support it. In order to run any commands
+        outside of a transaction use the no_transaction context manager.
+        """
+        if not os.path.exists(cls.database_path()):
+            # Make the database and protect it from snoopers and busybodies
+            con = sqlite3.connect(cls.database_path())
+            del con
+            os.chmod(cls.database_path(), 0o600)
+
+        con = sqlite3.connect(
+            cls.database_path(),
+            isolation_level="DEFERRED"
+        )
+
+        with cls.no_transaction(con):
+            # Turn on foreign keys.
+            # This has to be outside any transaction.
+            # See <https://stackoverflow.com/q/78898176>
+            con.execute("PRAGMA foreign_keys = ON")
+        # This has the side effect of definitely leaving autocommit off, which
+        # is what we want as the base state.
+
+        # Set up the connection to use the Row class so that we can look up row values by column name and not just order.
+        con.row_factory = sqlite3.Row
+        
+        return con
+
+    @classmethod
+    @contextmanager
+    def no_transaction(cls, con: sqlite3.Connection) -> Iterator[None]:
+        """
+        Temporarily disable the constant active transaction on the database
+        connection, on Python versions where it exists.
+
+        Commits the current transaction.
+        """
+        
+        con.commit()
+        if hasattr(con, 'autocommit'):
+            con.autocommit = True
+        yield
+        if hasattr(con, 'autocommit'):
+            con.autocommit = False
+
+    @classmethod
+    def ensure_tables(cls, con: sqlite3.Connection, cur: sqlite3.Cursor) -> None:
+        """
+        Ensure that tables exist in the database and the schema is migrated to the current version.
+
+        Leaves the cursor in a transaction where the schema version is known to be correct.
+
+        Must be called from inside a top-level methodf marked @db_retry.
+
+        :raises HistoryDatabaseSchemaTooNewError: If the schema is newer than the current version.
+        """
+
+        # Python already puts us in a transaction.
+
+        # TODO: Do a try-and-fall-back to avoid sending the table schema for
+        # this every time we do anything.
+        cur.execute("""
+            CREATE TABLE IF NOT EXISTS migrations (
+                version INT NOT NULL PRIMARY KEY,
+                description TEXT
+            )
+        """)
+        db_version = next(cur.execute("SELECT MAX(version) FROM migrations"))[0]
+        if db_version is None:
+            db_version = -1
+
+        # This holds pairs of description and command lists.
+        # To make a schema change, ADD A NEW PAIR AT THE END, and include
+        # statements to adjust existing data.
+        migrations = [
+            (
+                "Make initial tables",
+                [
+                    """
+                    CREATE TABLE workflows (
+                        id TEXT NOT NULL PRIMARY KEY,
+                        job_store TEXT NOT NULL,
+                        creation_time REAL NOT NULL,
+                        name TEXT,
+                        trs_spec TEXT
+                    )
+                    """,
+                    """
+                    CREATE INDEX idx_workflows_by_creation_time
+                    ON workflows (creation_time)
+                    """,
+                    # There's no reference constraint from the job attempts to
+                    # the workflow attempts because the jobs for a workflow
+                    # attempt need to go in before the attempt is known to be
+                    # finished or failed/before the attempt is submittable to
+                    # Dockstore.
+                    #
+                    # TODO: Should we force workflow attempts to be reported on
+                    # start so that we can have the jobs key-reference them?
+                    # And so that we always have a start time for the workflow
+                    # as a whole?
+                    """
+                    CREATE TABLE job_attempts (
+                        id TEXT NOT NULL PRIMARY KEY,
+                        workflow_id TEXT NOT NULL,
+                        workflow_attempt_number INT NOT NULL,
+                        job_name TEXT NOT NULL,
+                        succeeded INTEGER NOT NULL,
+                        start_time REAL NOT NULL,
+                        runtime REAL NOT NULL,
+                        cores REAL,
+                        cpu_seconds REAL,
+                        memory_bytes INTEGER,
+                        disk_bytes INTEGER,
+                        submitted_to_dockstore INTEGER NOT NULL DEFAULT FALSE,
+                        FOREIGN KEY(workflow_id) REFERENCES workflows(id)
+                    )
+                    """,
+                    """
+                    CREATE INDEX idx_job_attempts_by_workflow_attempt
+                    ON job_attempts (workflow_id, workflow_attempt_number)
+                    """,
+                    """
+                    CREATE TABLE workflow_attempts (
+                        workflow_id TEXT NOT NULL,
+                        attempt_number INTEGER NOT NULL,
+                        succeeded INTEGER NOT NULL,
+                        start_time REAL NOT NULL,
+                        runtime REAL NOT NULL,
+                        batch_system TEXT,
+                        caching INTEGER,
+                        toil_version TEXT,
+                        python_version TEXT,
+                        platform_system TEXT,
+                        platform_machine TEXT,
+                        submitted_to_dockstore INTEGER NOT NULL DEFAULT FALSE,
+                        PRIMARY KEY(workflow_id,attempt_number),
+                        FOREIGN KEY(workflow_id) REFERENCES workflows(id)
+                    )
+                    """
+                ],
+            ),
+        ]
+
+        if db_version + 1 > len(migrations):
+            raise HistoryDatabaseSchemaTooNewError(f"History database version is {db_version}, but known migrations only go up to {len(migrations) - 1}")
+
+        for migration_number in range(db_version + 1, len(migrations)):
+            for statement_number, statement in enumerate(migrations[migration_number][1]):
+                # Run all the migration commands.
+                # We don't use executescript() because (on old Pythons?) it
+                # commits the current transactrion first.
+                try:
+                    cur.execute(statement)
+                except sqlite3.OperationalError:
+                    logger.exception("Could not execute migration %s statement %s: %s", migration_number, statement_number, statement)
+                    raise
+            cur.execute("INSERT INTO migrations VALUES (?, ?)", (migration_number, migrations[migration_number][0]))
+
+        # If we did have to migrate, leave everything else we do as part of the migration transaction.
+
+    ##
+    # Recording Methods
+    ##
+
+    @classmethod
+    @db_retry
+    def record_workflow_creation(cls, workflow_id: str, job_store_spec: str) -> None:
+        """
+        Record that a workflow is being run.
+
+        Takes the Toil config's workflow ID and the location of the job store.
+
+        Should only be called on the *first* attempt on a job store, not on a
+        restart.
+
+        A workflow may have multiple attempts to run it, some of which succeed
+        and others of which fail. Probably only the last one should succeed.
+
+        :param job_store_spec: The job store specifier for the workflow. Should
+            be canonical and always start with the type and a colon. If the
+            job store is later moved by the user, the location will not be
+            updated.
+        """
+
+        if not cls.WORKFLOW_HISTORY_ENABLED:
+            return
+
+        logger.info("Recording workflow creation of %s in %s", workflow_id, job_store_spec)
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute("INSERT INTO workflows VALUES (?, ?, ?, NULL, NULL)", (workflow_id, job_store_spec, time.time()))
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        # If we raise out of here the connection goes away and the transaction rolls back.
+
+
+    @classmethod
+    @db_retry
+    def record_workflow_metadata(cls, workflow_id: str, workflow_name: str, trs_spec: Optional[str] = None) -> None:
+        """
+        Associate a name and optionally a TRS ID and version with a workflow run.
+        """
+
+        # TODO: Make name of this function less general?
+
+        if not cls.WORKFLOW_HISTORY_ENABLED:
+            return
+
+        logger.info("Workflow %s is a run of %s", workflow_id, workflow_name)
+        if trs_spec:
+            logger.info("Workflow %s has TRS ID and version %s", workflow_id, trs_spec)
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute("UPDATE workflows SET name = ? WHERE id = ?", (workflow_name, workflow_id))
+            if trs_spec is not None:
+                cur.execute("UPDATE workflows SET trs_spec = ? WHERE id = ?", (trs_spec, workflow_id))
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+    @classmethod
+    @db_retry
+    def record_job_attempt(
+            cls,
+            workflow_id: str,
+            workflow_attempt_number: int,
+            job_name: str,
+            succeeded: bool,
+            start_time: float,
+            runtime: float,
+            cores: Optional[float] = None,
+            cpu_seconds: Optional[float] = None,
+            memory_bytes: Optional[int] = None,
+            disk_bytes: Optional[int] = None
+        ) -> None:
+        """
+        Record that a job ran in a workflow.
+
+        Doesn't expect the provided information to uniquely identify the job
+        attempt; assigns the job attempt its own unique ID.
+
+        Thread safe.
+
+        :param job_name: A human-readable name for the job. Not expected to be
+            a job store ID or to necessarily uniquely identify the job within
+            the workflow.
+        :param start_time: Job execution start time ins econds since epoch.
+        :param runtime: Job execution duration in seconds.
+        :param cores: Number of CPU cores the job was scheduled on.
+        :param cpu_seconds: CPU core-seconds actually consumed.
+        :param memory_bytes: Peak observed job memory usage.
+        :param disk_bytes: Observed job disk usage.
+        """
+
+        if not cls.WORKFLOW_HISTORY_ENABLED or not cls.JOB_HISTORY_ENABLED:
+            return
+
+        logger.debug("Workflow %s ran job %s", workflow_id, job_name)
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                INSERT INTO job_attempts(
+                    id,
+                    workflow_id,
+                    workflow_attempt_number,
+                    job_name,
+                    succeeded,
+                    start_time,
+                    runtime,
+                    cores,
+                    cpu_seconds,
+                    memory_bytes,
+                    disk_bytes
+                )
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+                """,
+                (
+                    str(uuid.uuid4()),
+                    workflow_id,
+                    workflow_attempt_number,
+                    job_name,
+                    1 if succeeded else 0,
+                    start_time,
+                    runtime,
+                    cores,
+                    cpu_seconds,
+                    memory_bytes,
+                    disk_bytes,
+                )
+            )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+    @classmethod
+    @db_retry
+    def record_workflow_attempt(
+        cls,
+        workflow_id: str,
+        workflow_attempt_number: int,
+        succeeded: bool,
+        start_time: float,
+        runtime: float,
+        batch_system: Optional[str] = None,
+        caching: Optional[bool] = None,
+        toil_version: Optional[str] = None,
+        python_version: Optional[str] = None,
+        platform_system: Optional[str] = None,
+        platform_machine: Optional[str] = None
+    ) -> None:
+        """
+        Record a workflow attempt (start or restart) having finished or failed.
+
+        :param batch_system: The Python type name of the batch system implementation used.
+        :param caching: Whether Toil filestore-level caching was used.
+        :param toil_version: Version of Toil used to run the workflow.
+        :param python_version: Version of Python used to run the workflow.
+        :param platform_system: OS ("Darwin", "Linux", etc.) used to run the workflow.
+        :param platform_machine: CPU type ("AMD64", etc.) used to run the workflow leader.
+        """
+
+        if not cls.WORKFLOW_HISTORY_ENABLED:
+            return
+
+        logger.info("Workflow %s stopped. Success: %s", workflow_id, succeeded)
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                INSERT INTO workflow_attempts(
+                    workflow_id,
+                    attempt_number,
+                    succeeded,
+                    start_time,
+                    runtime,
+                    batch_system,
+                    caching,
+                    toil_version,
+                    python_version,
+                    platform_system,
+                    platform_machine
+                )
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+                """,
+                (
+                    workflow_id,
+                    workflow_attempt_number,
+                    1 if succeeded else 0,
+                    start_time,
+                    runtime,
+                    batch_system,
+                    caching,
+                    toil_version,
+                    python_version,
+                    platform_system,
+                    platform_machine
+                )
+            )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+
+    ##
+    # Read methods
+    ##
+
+    # We would implement a bunch of iterators and allow follow-up queries, but
+    # then we'd have to figure out how to make sure we use one connection and
+    # cursor and not block ourselves with the database transaction locks.
+    #
+    # So instead we always fetch all the information asked for and close out
+    # the read transaction before returning.
+    #
+    # This means the caller has to worry about a workflow vanishing or changing
+    # between when it was shown to them and when they ask follow-up questions,
+    # but it also means we can't deadlock.
+
+    @classmethod
+    @db_retry
+    def summarize_workflows(cls) -> list[WorkflowSummary]:
+        """
+        List all known workflows and their summary statistics.
+        """
+
+        workflows = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                SELECT
+                    workflows.id AS id,
+                    workflows.name AS name,
+                    workflows.job_store AS job_store,
+                    (SELECT count(*) FROM workflow_attempts WHERE workflow_id = workflows.id) AS total_attempts,
+                    (SELECT count(*) FROM job_attempts WHERE workflow_id = workflows.id) AS total_job_attempts,
+                    (SELECT min(count(*), 1) FROM workflow_attempts WHERE workflow_id = workflows.id AND succeeded = TRUE) AS succeeded,
+                    (SELECT min(start_time) FROM workflow_attempts WHERE workflow_id = workflows.id) AS start_time,
+                    (SELECT max(start_time + runtime) FROM workflow_attempts WHERE workflow_id = workflows.id) AS end_time,
+                    workflows.trs_spec AS trs_spec
+                FROM workflows
+                ORDER BY start_time DESC
+                """
+            )
+            for row in cur:
+                workflows.append(
+                    WorkflowSummary(
+                        id=row["id"],
+                        name=row["name"],
+                        job_store=row["job_store"],
+                        total_attempts=row["total_attempts"],
+                        total_job_attempts=row["total_job_attempts"],
+                        succeeded=(row["succeeded"] == 1),
+                        start_time=row["start_time"],
+                        runtime=(row["end_time"] - row["start_time"]) if row["start_time"] is not None and row["end_time"] is not None else None,
+                        trs_spec=row["trs_spec"]
+                    )
+                )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return workflows
+
+    @classmethod
+    @db_retry
+    def get_submittable_workflow_attempts(cls, limit: int = sys.maxsize) -> list[WorkflowAttemptSummary]:
+        """
+        List all workflow attempts not yet submitted to Dockstore.
+
+        :param limit: Get no more than this many.
+        """
+
+        attempts = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                SELECT
+                    workflow_attempts.workflow_id AS workflow_id,
+                    workflow_attempts.attempt_number AS attempt_number,
+                    workflow_attempts.succeeded AS succeeded,
+                    workflow_attempts.start_time AS start_time,
+                    workflow_attempts.runtime AS runtime,
+                    workflow_attempts.batch_system AS batch_system,
+                    workflow_attempts.caching AS caching,
+                    workflow_attempts.toil_version AS toil_version,
+                    workflow_attempts.python_version AS python_version,
+                    workflow_attempts.platform_system AS platform_system,
+                    workflow_attempts.platform_machine AS platform_machine,
+                    workflow_attempts.submitted_to_dockstore AS submitted_to_dockstore,
+                    workflows.job_store AS workflow_job_store,
+                    workflows.trs_spec AS workflow_trs_spec
+                FROM workflow_attempts
+                    JOIN workflows ON workflow_attempts.workflow_id = workflows.id
+                WHERE workflow_attempts.submitted_to_dockstore = FALSE
+                    AND workflows.trs_spec IS NOT NULL
+                ORDER BY start_time DESC
+                LIMIT ?
+                """,
+                (limit,)
+            )
+            for row in cur:
+                attempts.append(
+                    WorkflowAttemptSummary(
+                        workflow_id=row["workflow_id"],
+                        attempt_number=row["attempt_number"],
+                        succeeded=(row["succeeded"] == 1),
+                        start_time=row["start_time"],
+                        runtime=row["runtime"],
+                        batch_system=row["batch_system"],
+                        caching=(row["caching"] == 1),
+                        toil_version=row["toil_version"],
+                        python_version=row["python_version"],
+                        platform_system=row["platform_system"],
+                        platform_machine=row["platform_machine"],
+                        submitted_to_dockstore=(row["submitted_to_dockstore"] == 1),
+                        workflow_job_store=row["workflow_job_store"],
+                        workflow_trs_spec=row["workflow_trs_spec"]
+                    )
+                )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return attempts
+
+    @classmethod
+    @db_retry
+    def get_workflow_attempts_with_submittable_job_attempts(cls, limit: int = sys.maxsize) -> list[WorkflowAttemptSummary]:
+        """
+        Get all workflow attempts that have job attempts not yet submitted to
+        Dockstore.
+
+        The workflow attempts themselves will have finished and been recorded,
+        and have TRS IDs.
+
+        :param limit: Get no more than this many.
+        """
+
+        attempts = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                SELECT
+                    workflow_attempts.workflow_id AS workflow_id,
+                    workflow_attempts.attempt_number AS attempt_number,
+                    workflow_attempts.succeeded AS succeeded,
+                    workflow_attempts.start_time AS start_time,
+                    workflow_attempts.runtime AS runtime,
+                    workflow_attempts.batch_system AS batch_system,
+                    workflow_attempts.caching AS caching,
+                    workflow_attempts.toil_version AS toil_version,
+                    workflow_attempts.python_version AS python_version,
+                    workflow_attempts.platform_system AS platform_system,
+                    workflow_attempts.platform_machine AS platform_machine,
+                    workflow_attempts.submitted_to_dockstore AS submitted_to_dockstore,
+                    workflows.job_store AS workflow_job_store,
+                    workflows.trs_spec AS workflow_trs_spec
+                FROM (
+                    SELECT DISTINCT
+                        workflow_id, workflow_attempt_number
+                    FROM job_attempts
+                    WHERE job_attempts.submitted_to_dockstore = FALSE
+                ) AS found_job_attempts
+                    JOIN workflows ON found_job_attempts.workflow_id = workflows.id
+                    JOIN workflow_attempts ON
+                        found_job_attempts.workflow_id = workflow_attempts.workflow_id
+                        AND found_job_attempts.workflow_attempt_number = workflow_attempts.attempt_number
+                WHERE workflows.trs_spec IS NOT NULL
+                LIMIT ?
+                """,
+                (limit,)
+            )
+            for row in cur:
+                # TODO: Unify row to data class conversion
+                attempts.append(
+                    WorkflowAttemptSummary(
+                        workflow_id=row["workflow_id"],
+                        attempt_number=row["attempt_number"],
+                        succeeded=(row["succeeded"] == 1),
+                        start_time=row["start_time"],
+                        runtime=row["runtime"],
+                        batch_system=row["batch_system"],
+                        caching=(row["caching"] == 1),
+                        toil_version=row["toil_version"],
+                        python_version=row["python_version"],
+                        platform_system=row["platform_system"],
+                        platform_machine=row["platform_machine"],
+                        submitted_to_dockstore=(row["submitted_to_dockstore"] == 1),
+                        workflow_job_store=row["workflow_job_store"],
+                        workflow_trs_spec=row["workflow_trs_spec"]
+                    )
+                )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return attempts
+
+    @classmethod
+    @db_retry
+    def get_workflow_attempt(cls, workflow_id: str, attempt_number: int) -> Optional[WorkflowAttemptSummary]:
+        """
+        Get a single (not necessarily unsubmitted, not necessarily TRS-ID-having) workflow attempt summary, if present.
+        """
+
+        # TODO: Consolidate with the other 2 ways to query workflow attempts!
+
+        attempts = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                SELECT
+                    workflow_attempts.workflow_id AS workflow_id,
+                    workflow_attempts.attempt_number AS attempt_number,
+                    workflow_attempts.succeeded AS succeeded,
+                    workflow_attempts.start_time AS start_time,
+                    workflow_attempts.runtime AS runtime,
+                    workflow_attempts.batch_system AS batch_system,
+                    workflow_attempts.caching AS caching,
+                    workflow_attempts.toil_version AS toil_version,
+                    workflow_attempts.python_version AS python_version,
+                    workflow_attempts.platform_system AS platform_system,
+                    workflow_attempts.platform_machine AS platform_machine,
+                    workflow_attempts.submitted_to_dockstore AS submitted_to_dockstore,
+                    workflows.job_store AS workflow_job_store,
+                    workflows.trs_spec AS workflow_trs_spec
+                FROM workflow_attempts
+                    JOIN workflows ON workflow_attempts.workflow_id = workflows.id
+                WHERE workflow_id = ?
+                    AND attempt_number = ?
+                ORDER BY start_time DESC
+                LIMIT 1
+                """,
+                (workflow_id, attempt_number)
+            )
+            for row in cur:
+                attempts.append(
+                    WorkflowAttemptSummary(
+                        workflow_id=row["workflow_id"],
+                        attempt_number=row["attempt_number"],
+                        succeeded=(row["succeeded"] == 1),
+                        start_time=row["start_time"],
+                        runtime=row["runtime"],
+                        batch_system=row["batch_system"],
+                        caching=(row["caching"] == 1),
+                        toil_version=row["toil_version"],
+                        python_version=row["python_version"],
+                        platform_system=row["platform_system"],
+                        platform_machine=row["platform_machine"],
+                        submitted_to_dockstore=(row["submitted_to_dockstore"] == 1),
+                        workflow_job_store=row["workflow_job_store"],
+                        workflow_trs_spec=row["workflow_trs_spec"]
+                    )
+                )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        if len(attempts) == 0:
+            # Not found
+            return None
+        else:
+            return attempts[0]
+
+    @classmethod
+    @db_retry
+    def get_unsubmitted_job_attempts(cls, workflow_id: str, attempt_number: int) -> list[JobAttemptSummary]:
+        """
+        List all job attempts in the given workflow attempt not yet submitted to Dockstore.
+
+        Doesn't check to make sure the workflow has a TRS ID.
+        """
+
+        attempts = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                """
+                SELECT
+                    id,
+                    job_name,
+                    succeeded,
+                    start_time,
+                    runtime,
+                    cores,
+                    cpu_seconds,
+                    memory_bytes,
+                    disk_bytes,
+                    submitted_to_dockstore
+                FROM job_attempts
+                WHERE workflow_id = ?
+                AND workflow_attempt_number = ?
+                AND submitted_to_dockstore = FALSE
+                ORDER BY start_time DESC
+                """,
+                (workflow_id, attempt_number)
+            )
+            for row in cur:
+                attempts.append(
+                    JobAttemptSummary(
+                        id=row["id"],
+                        job_name=row["job_name"],
+                        succeeded=(row["succeeded"] == 1),
+                        start_time=row["start_time"],
+                        runtime=row["runtime"],
+                        cores=row["cores"],
+                        cpu_seconds=row["cpu_seconds"],
+                        memory_bytes=row["memory_bytes"],
+                        disk_bytes=row["disk_bytes"],
+                        submitted_to_dockstore=(row["submitted_to_dockstore"] == 1)
+                    )
+                )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return attempts
+
+    ###
+    # Submission marking methods
+    ###
+
+    @classmethod
+    @db_retry
+    def mark_workflow_attempt_submitted(cls, workflow_id: str, attempt_number: int) -> None:
+        """
+        Mark a workflow attempt as having been successfully submitted to Dockstore.
+
+        Does not mark the workflow attempt's job attempts as submitted.
+        """
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            cur.execute(
+                "UPDATE workflow_attempts SET submitted_to_dockstore = TRUE WHERE workflow_id = ? AND attempt_number = ?",
+                (workflow_id, attempt_number)
+            )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+    @classmethod
+    @db_retry
+    def mark_job_attempts_submitted(cls, job_attempt_ids: list[str]) -> None:
+        """
+        Mark a collection of job attempts as submitted to Dockstore in a single transaction.
+        """
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+            for job_attempt_id in job_attempt_ids:
+                # Do all the marking in one transaction
+                cur.execute(
+                    "UPDATE job_attempts SET submitted_to_dockstore = TRUE WHERE id = ?",
+                    (job_attempt_id,)
+                )
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+    @classmethod
+    @db_retry
+    def count_workflows(cls) -> int:
+        """
+        Count workflows in the database.
+        """
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            cur.execute("SELECT count(*) FROM workflows")
+
+            count = cur.fetchone()[0]
+            assert isinstance(count, int)
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return count
+
+    @classmethod
+    @db_retry
+    def count_workflow_attempts(cls) -> int:
+        """
+        Count workflow attempts in the database.
+        """
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            cur.execute("SELECT count(*) FROM workflow_attempts")
+
+            count = cur.fetchone()[0]
+            assert isinstance(count, int)
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return count
+
+    @classmethod
+    @db_retry
+    def count_job_attempts(cls) -> int:
+        """
+        Count job attempts in the database.
+        """
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            cur.execute("SELECT count(*) FROM job_attempts")
+
+            count = cur.fetchone()[0]
+            assert isinstance(count, int)
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return count
+
+    @classmethod
+    @db_retry
+    def get_fully_submitted_workflow_ids(cls, limit: int = sys.maxsize) -> list[str]:
+        """
+        Get workflows that have a successful attempt and no unsubmitted attempts or job attempts.
+        """
+        ids = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            cur.execute(
+                """
+                SELECT
+                    workflows.id
+                FROM workflows
+                WHERE
+                    (
+                        SELECT
+                            count(*)
+                        FROM workflow_attempts
+                        WHERE workflow_id = workflows.id
+                            AND succeeded = TRUE
+                            AND submitted_to_dockstore = TRUE
+                        LIMIT 1
+                    ) = 1
+                    AND (
+                        SELECT
+                            count(*)
+                        FROM workflow_attempts
+                        WHERE workflow_id = workflows.id
+                            AND submitted_to_dockstore = FALSE
+                        LIMIT 1
+                    ) = 0
+                    AND (
+                        SELECT
+                            count(*)
+                        FROM job_attempts
+                        WHERE workflow_id = workflows.id
+                            AND submitted_to_dockstore = FALSE
+                        LIMIT 1
+                    ) = 0
+                LIMIT ?
+                """,
+                (limit,)
+            )
+            for row in cur:
+                ids.append(row["id"])
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return ids
+
+    @classmethod
+    @db_retry
+    def get_oldest_workflow_ids(cls, limit: int = sys.maxsize) -> list[str]:
+        """
+        Get workflows that are old.
+        """
+
+        ids = []
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            # We could use a complicated query to bump workflows down the list
+            # if they have been updated by having attempts or job attempts. But
+            # that would mean we'd need to do a lot of querying and live
+            # sorting, whereas using just the creation time lets us use an
+            # index and a limit efficiently.
+
+            cur.execute(
+                """
+                SELECT
+                    id,
+                    creation_time
+                FROM workflows
+                ORDER BY creation_time ASC
+                LIMIT ?
+                """,
+                (limit,)
+            )
+            for row in cur:
+                ids.append(row["id"])
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return ids
+
+    @classmethod
+    @db_retry
+    def delete_workflow(cls, workflow_id: str) -> None:
+        """
+        Delete a workflow and all its attempts and job attempts.
+
+        Succeeds if the workflow does not exist.
+        """
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            cur.execute("DELETE FROM job_attempts WHERE workflow_id = ?", (workflow_id,))
+            cur.execute("DELETE FROM workflow_attempts WHERE workflow_id = ?", (workflow_id,))
+            cur.execute("DELETE FROM workflows WHERE id = ?", (workflow_id,))
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+    @classmethod
+    @db_retry
+    def get_database_byte_size(cls) -> int:
+        """
+        Get the total number of bytes used by the database.
+        """
+
+        con = cls.connection()
+        cur = con.cursor()
+        try:
+            cls.ensure_tables(con, cur)
+
+            cur.execute("PRAGMA page_size")
+            page_size = cur.fetchone()[0]
+            assert isinstance(page_size, int)
+
+            cur.execute("PRAGMA page_count")
+            page_count = cur.fetchone()[0]
+            assert isinstance(page_count, int)
+
+        except:
+            con.rollback()
+            con.close()
+            raise
+        else:
+            con.commit()
+            con.close()
+
+        return page_size * page_count
+
+    @classmethod
+    @db_retry
+    def compact_database(cls) -> None:
+        """
+        Shrink the database to remove unused space.
+        """
+        
+        con = cls.connection()
+        cur = con.cursor()
+
+        # Don't bother making tables; we don't need them for this and they need
+        # a transaction.
+        
+        with cls.no_transaction(con):
+            # Do the vacuum outside any transaction, and rely on it to
+            # synchronize appropriately internally.
+            cur.execute("VACUUM")
+
+        con.close()
+
+    @classmethod
+    def enforce_byte_size_limit(cls, limit: int = 100 * 1024 * 1024) -> None:
+        """
+        Shrink the database until it is smaller than the given limit, or until
+        it is empty, by throwing away workflows.
+
+        Throws data away in a sensible order, least important to most
+        important.
+        """
+
+        db_size = cls.get_database_byte_size()
+
+        if db_size < limit:
+            # Nothing to do!
+            return
+
+        while db_size > limit:
+            # Look for some things we submitted already
+            target_workflows = cls.get_fully_submitted_workflow_ids(limit=100)
+            if len(target_workflows) == 0:
+                # If there aren't any, do oldest workflows a few at a time
+                # We need to balance the O(n^2)
+                # delete-and-copy-the-whole-db-to-vacuum loop with not wanting
+                # to delete too many workflows we could keep.
+                target_workflows = cls.get_oldest_workflow_ids(limit=10)
+            if len(target_workflows) == 0:
+                # There are no more workflows to delete.
+                break
+
+            for workflow_id in target_workflows:
+                # Delete all the workflows we don't want.
+                cls.delete_workflow(workflow_id)
+
+            # Shrink the DB
+            cls.compact_database()
+            # Re-check the size
+            db_size = cls.get_database_byte_size()
+
+
+
+
+    @classmethod
+    def database_dump_lines(cls) -> Iterable[str]:
+        """
+        Yield lines from the database dump.
+
+        For debugging tests.
+        """
+        return cls.connection().iterdump()
+
+
+
+