alembic-runner 0.1.0__py3-none-any.whl

alembic_runner/__init__.py

@@ -0,0 +1,61 @@
+"""run alembic plan, review and apply from a host application."""
+
+from .backend import CONFIG_KEYS, KINDS, TOKEN_ENV, Backend, BackendError, StateStore, token_env
+from .plan import ApplyReport, DocumentError, DriftReport, Plan
+from .runner import (
+    SUPPORTED,
+    ApplyOutcome,
+    Completed,
+    DriftOutcome,
+    Flow,
+    InventoryOutcome,
+    PlanOutcome,
+    Runner,
+    RunnerError,
+    StaleCheck,
+)
+from .status import (
+    ACTIVE,
+    TERMINAL,
+    TRANSITIONS,
+    RunStatus,
+    TransitionError,
+    check_transition,
+    failure_status,
+)
+from .workspace import Workspace, WorkspaceError
+
+__version__ = "0.1.0"
+
+__all__ = [
+    "ACTIVE",
+    "CONFIG_KEYS",
+    "KINDS",
+    "SUPPORTED",
+    "TERMINAL",
+    "TOKEN_ENV",
+    "TRANSITIONS",
+    "ApplyOutcome",
+    "ApplyReport",
+    "Backend",
+    "BackendError",
+    "Completed",
+    "DocumentError",
+    "DriftOutcome",
+    "DriftReport",
+    "Flow",
+    "InventoryOutcome",
+    "Plan",
+    "PlanOutcome",
+    "RunStatus",
+    "Runner",
+    "RunnerError",
+    "StaleCheck",
+    "StateStore",
+    "TransitionError",
+    "Workspace",
+    "WorkspaceError",
+    "check_transition",
+    "failure_status",
+    "token_env",
+]

alembic_runner/backend.py

@@ -0,0 +1,111 @@
+"""backends and state stores, as the environment and files a run hands to alembic."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+KINDS = ("netbox", "nautobot", "infrahub", "peeringdb", "external")
+
+# the config keys a stored backend may set. anything that names a command, a
+# working directory or an environment for a process stays in host settings:
+# whoever can edit a backend must not be able to run something on the worker.
+CONFIG_KEYS = {
+    "netbox": frozenset({"url", "instance"}),
+    "nautobot": frozenset({"url", "instance"}),
+    "infrahub": frozenset({"url", "instance"}),
+    "peeringdb": frozenset({"url", "instance"}),
+    "external": frozenset({"args", "setup", "timeout_seconds", "instance"}),
+}
+
+# the environment variable each built-in adapter reads its credential from.
+TOKEN_ENV = {
+    "netbox": "NETBOX_TOKEN",
+    "nautobot": "NAUTOBOT_TOKEN",
+    "infrahub": "INFRAHUB_TOKEN",
+    "peeringdb": "PEERINGDB_API_KEY",
+}
+
+
+class BackendError(ValueError):
+    pass
+
+
+@dataclass(frozen=True)
+class Backend:
+    """one backend as a run sees it.
+
+    `config` is the non-secret part of an alembic backend config, without the
+    `backend:` key. `env` holds the credentials, by variable name. an external
+    backend's `command` comes from host settings, never from stored config.
+    """
+
+    kind: str
+    config: dict[str, Any] = field(default_factory=dict)
+    env: dict[str, str] = field(default_factory=dict)
+    command: str | None = None
+
+    def __post_init__(self) -> None:
+        if self.kind not in KINDS:
+            raise BackendError(f"unknown backend kind {self.kind!r}")
+        unknown = self.config.keys() - CONFIG_KEYS[self.kind]
+        if unknown:
+            raise BackendError(
+                f"a {self.kind} backend config cannot set {', '.join(sorted(unknown))}; "
+                f"it takes {', '.join(sorted(CONFIG_KEYS[self.kind]))}"
+            )
+        if self.kind == "external":
+            if not self.command:
+                raise BackendError("an external backend needs a command from host settings")
+        elif self.command:
+            raise BackendError(f"a {self.kind} backend takes no command")
+
+    def config_document(self) -> dict[str, Any]:
+        doc = {"backend": self.kind, **self.config}
+        if self.command:
+            doc["command"] = self.command
+        return doc
+
+
+def token_env(kind: str, token: str) -> dict[str, str]:
+    """the environment that hands `token` to a built-in adapter of `kind`."""
+    try:
+        return {TOKEN_ENV[kind]: token}
+    except KeyError:
+        raise BackendError(f"a {kind} backend reads no token variable") from None
+
+
+@dataclass(frozen=True)
+class StateStore:
+    """where identity state lives. `local` keeps it in the flow directory."""
+
+    backend: str = "local"
+    postgres_url: str | None = None
+    postgres_tls: str | None = None
+
+    def __post_init__(self) -> None:
+        if self.backend not in ("local", "postgres"):
+            raise BackendError(f"unknown state backend {self.backend!r}")
+        if self.backend == "postgres" and not self.postgres_url:
+            raise BackendError("postgres state needs a postgres_url")
+
+    @classmethod
+    def from_settings(cls, settings: dict[str, Any] | None) -> StateStore:
+        settings = settings or {}
+        return cls(
+            backend=settings.get("backend", "local"),
+            postgres_url=settings.get("postgres_url"),
+            postgres_tls=settings.get("postgres_tls"),
+        )
+
+    def env(self, key: str) -> dict[str, str]:
+        if self.backend == "local":
+            return {"ALEMBIC_STATE_BACKEND": "local"}
+        env = {
+            "ALEMBIC_STATE_BACKEND": "postgres",
+            "ALEMBIC_STATE_POSTGRES_URL": self.postgres_url or "",
+            "ALEMBIC_STATE_KEY": key,
+        }
+        if self.postgres_tls:
+            env["ALEMBIC_STATE_POSTGRES_TLS"] = self.postgres_tls
+        return env

alembic_runner/hosting.py

@@ -0,0 +1,197 @@
+"""run bookkeeping the host apps share: moves, locks, approval checks and the job
+wrapper. it needs django, so only the host apps import it; the rest of the core
+stays stdlib-only.
+
+a host's run model carries the fields both apps define: `flow`, `kind`,
+`status`, `requested_by`, `decided_by`, `decided_at`, `planned_at`,
+`apply_started_at`, `finished_at`, `plan`, `plan_sha256`, `approved_sha256`,
+`error`, `output`."""
+
+from __future__ import annotations
+
+from collections.abc import Callable, Iterable
+from datetime import timedelta
+from typing import Any
+
+from django.core.exceptions import PermissionDenied, ValidationError
+from django.db import IntegrityError, transaction
+from django.utils import timezone
+
+from .pipeline import Outcome, RunFailed, approved_plan
+from .plan import Plan
+from .status import RunStatus, TransitionError, check_transition, failure_status
+
+S = RunStatus
+
+# a new plan of a flow takes these runs' place.
+SUPERSEDABLE = (S.AWAITING_APPROVAL.value, S.APPLY_FAILED.value)
+
+
+class ActionError(ValidationError):
+    """a run cannot do what was asked of it in its current state."""
+
+
+def _save(run) -> None:
+    # nautobot validates on save through `validated_save`; netbox has only `save`.
+    getattr(run, "validated_save", run.save)()
+
+
+def move(run, new: RunStatus | str, **fields: Any):
+    """move a locked run to `new`, setting `fields`, or raise ActionError."""
+    try:
+        check_transition(run.status, new)
+    except TransitionError as e:
+        raise ActionError(str(e)) from e
+    run.status = RunStatus(new).value
+    for name, value in fields.items():
+        setattr(run, name, value)
+    _save(run)
+    return run
+
+
+def locked(run):
+    return type(run).objects.select_for_update().get(pk=run.pk)
+
+
+def expired(run, ttl_hours: int | None) -> bool:
+    return bool(
+        ttl_hours
+        and run.planned_at
+        and timezone.now() > run.planned_at + timedelta(hours=ttl_hours)
+    )
+
+
+def create_run(model, flow, user, kind: str, supersede: bool, retire: Callable = lambda run: None):
+    """create a pending run of `flow`. a plan (`supersede`) takes the place of the
+    flow's waiting runs, each passed to `retire` first. call inside a transaction."""
+    if supersede:
+        for run in model.objects.select_for_update().filter(flow=flow, status__in=SUPERSEDABLE):
+            retire(run)
+            move(run, S.SUPERSEDED, finished_at=timezone.now())
+    try:
+        with transaction.atomic():
+            return model.objects.create(flow=flow, kind=kind, requested_by=user)
+    except IntegrityError:
+        raise ActionError("This flow already has a run in progress.") from None
+
+
+def _approve_perm(run) -> str:
+    return f"{run._meta.app_label}.approve_run"
+
+
+def refusal(run, user, distinct: bool, ttl_hours: int | None = None) -> str | None:
+    """why `user` may not approve or reject `run`, or None if they may."""
+    if not user.has_perm(_approve_perm(run), run):
+        return "You may not approve or reject this run."
+    if distinct and run.requested_by_id == user.pk:
+        return "Someone other than the requester approves this run."
+    if expired(run, ttl_hours):
+        return "This plan is too old to approve. Plan again."
+    return None
+
+
+def check_approver(run, user, distinct: bool) -> None:
+    reason = refusal(run, user, distinct)
+    if reason:
+        raise PermissionDenied(reason)
+
+
+def approve(run, user, ttl_hours: int | None) -> bool:
+    """move a locked, waiting run to approved, or to expired when its plan is too
+    old. returns whether it was approved; the caller queues the apply."""
+    if expired(run, ttl_hours):
+        move(run, S.EXPIRED, finished_at=timezone.now())
+        return False
+    move(
+        run,
+        S.APPROVED,
+        decided_by=user,
+        decided_at=timezone.now(),
+        approved_sha256=run.plan_sha256,
+    )
+    return True
+
+
+def reject(run, user):
+    now = timezone.now()
+    return move(run, S.REJECTED, decided_by=user, decided_at=now, finished_at=now)
+
+
+def check_resumable(run, user) -> None:
+    if not user.has_perm(_approve_perm(run), run):
+        raise PermissionDenied("You may not resume this run.")
+    if run.status != S.APPLY_FAILED.value:
+        raise ActionError("Only a failed apply can be resumed.")
+
+
+# jobs
+
+
+def transition(run, new: RunStatus | str, **fields: Any) -> None:
+    with transaction.atomic():
+        move(locked(run), new, **fields)
+    run.refresh_from_db()
+
+
+def settle(run, outcome: Outcome, **fields: Any) -> None:
+    """store what a pipeline step decided on the run."""
+    if outcome.finished:
+        fields["finished_at"] = timezone.now()
+    transition(run, outcome.status, **outcome.fields, **fields)
+
+
+def fail(run, message: str, output: str = "") -> None:
+    """record a failure on the run, in the status its phase fails into."""
+    fields: dict[str, Any] = {"error": message}
+    if output:
+        fields["output"] = output
+    with transaction.atomic():
+        run = locked(run)
+        new = failure_status(run.status)
+        if new is S.FAILED:
+            fields["finished_at"] = timezone.now()
+        if new is not None:
+            move(run, new, **fields)
+
+
+def run_guarded(run, execute: Callable, logger, known: Iterable[type[Exception]] = ()) -> None:
+    """run a job's body. a failure it expects becomes the run's status; anything
+    else does too, and is raised again so the host marks its job as errored."""
+    try:
+        execute(run)
+    except RunFailed as e:
+        logger.error(str(e))
+        fail(run, str(e), e.output)
+    except ActionError as e:
+        # the run is not in a state this job can act on; leave it as it is.
+        logger.error(" ".join(e.messages))
+    except tuple(known) as e:
+        logger.error(str(e))
+        fail(run, str(e))
+    except Exception as e:
+        fail(run, f"{type(e).__name__}: {e}")
+        raise
+
+
+def start_apply(run) -> tuple[Plan, bool]:
+    """take the run into applying: one apply per target at a time, from an
+    approved run or a failed apply, and only with the plan that was approved.
+    returns the plan and whether this resumes a failed apply."""
+    model = type(run)
+    with transaction.atomic():
+        current = locked(run)
+        busy = (
+            model.objects.filter(flow__target=run.flow.target, status=S.APPLYING.value)
+            .exclude(pk=run.pk)
+            .select_for_update()
+        )
+        if busy.exists():
+            raise RunFailed(f"Another run is applying to {run.flow.target}; try again later.")
+        if current.status not in (S.APPROVED.value, S.APPLY_FAILED.value):
+            raise ActionError(f"a {current.status} run is not applied")
+        resuming = current.status == S.APPLY_FAILED.value
+        # checked before the move, so a plan that fails it never becomes resumable.
+        plan = approved_plan(current.plan, current.approved_sha256)
+        move(current, S.APPLYING, apply_started_at=current.apply_started_at or timezone.now())
+    run.refresh_from_db()
+    return plan, resuming

alembic_runner/pipeline.py

@@ -0,0 +1,224 @@
+"""what a plan job and an apply job do, apart from where the host keeps the run.
+
+the host snapshots the flow's files, locks and moves the run, and stores the
+fields these functions return; everything alembic-shaped happens here."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import logging
+from collections.abc import Iterable
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+from .plan import Plan
+from .runner import Completed, Flow, Runner
+from .status import RunStatus
+
+S = RunStatus
+
+SNAPSHOT_LIMIT = 20 << 20
+
+# what a run keeps when an import or a map produced its inventory.
+DERIVED_ENTRY = "inventory.json"
+
+
+class RunFailed(Exception):
+    """a run that cannot go on, with the alembic output that led there."""
+
+    def __init__(self, message: str, output: str = ""):
+        super().__init__(message)
+        self.output = output
+
+
+def failed(outcome: Any, output: str) -> RunFailed:
+    return RunFailed(f"{outcome.error}\n{outcome.completed.stderr}".strip(), output)
+
+
+def output_text(*completed: Completed, before: str = "") -> str:
+    """alembic's output, one block per command, after what earlier jobs recorded."""
+    blocks = [before.rstrip()] if before else []
+    for c in completed:
+        mode = next((f" {flag}" for flag in ("--dry-run", "--report") if flag in c.argv), "")
+        head = f"$ alembic {c.argv[1]}{mode}"
+        if mode == " --dry-run":
+            head += "  (stale check)"
+        body = f"{c.stdout}{c.stderr}".rstrip()
+        blocks.append(f"{head}\n{body}" if body else head)
+    return "\n\n".join(blocks)
+
+
+def input_sha256(files: dict[str, str]) -> str:
+    return hashlib.sha256(json.dumps(files, sort_keys=True).encode()).hexdigest()
+
+
+def snapshot(
+    items: Iterable[tuple[str, bytes]],
+    entry: str,
+    where: str,
+    logger: logging.Logger,
+) -> dict[str, str]:
+    """a flow's files, from `(path under the root, bytes)` pairs. binary files are
+    skipped, the total is capped, and `entry` must be among them."""
+    files, total = {}, 0
+    for path, data in items:
+        total += len(data)
+        if total > SNAPSHOT_LIMIT:
+            raise RunFailed(f"The files in {where} exceed {SNAPSHOT_LIMIT} bytes.")
+        try:
+            files[path] = data.decode("utf-8")
+        except UnicodeDecodeError:
+            logger.info(f"skipping {path}: not utf-8 text")
+    if entry not in files:
+        raise RunFailed(f"{entry} is not in {where}.")
+    return files
+
+
+def derive(
+    runner: Runner,
+    flow: Flow,
+    run: str,
+    files: dict[str, str],
+    entry: str,
+    map_spec: str,
+    where: str,
+    logger: logging.Logger,
+) -> tuple[dict[str, str], str, list[Completed]]:
+    """import from the flow's source and apply its map spec, when it has them.
+    returns the files to plan from, the entry among them, and what ran."""
+    inventory = runner.workspace(flow).write_input(run, files, entry)
+    completed: list[Completed] = []
+    current = inventory
+    if flow.source is not None:
+        logger.info("importing from the flow's source")
+        imported = runner.import_(flow, run, inventory)
+        completed.append(imported.completed)
+        if imported.error:
+            raise failed(imported, output_text(*completed))
+        current = imported.inventory
+    if map_spec:
+        if map_spec not in files:
+            raise RunFailed(f"map spec {map_spec} is not in {where}")
+        spec = runner.workspace(flow).input_dir(run) / map_spec
+        mapped = runner.map(flow, run, current, spec)
+        completed.append(mapped.completed)
+        if mapped.error:
+            raise failed(mapped, output_text(*completed))
+        current = mapped.inventory
+    if current == inventory:
+        return files, entry, completed
+    # the plan, the stale check and the apply all work from the derived
+    # inventory, so that is what the run keeps.
+    return {DERIVED_ENTRY: Path(current).read_text(encoding="utf-8")}, DERIVED_ENTRY, completed
+
+
+@dataclass(frozen=True)
+class Outcome:
+    """where a job leaves the run, and what the host stores on it."""
+
+    status: RunStatus
+    fields: dict[str, Any] = field(default_factory=dict)
+    finished: bool = True
+
+
+def plan_run(
+    runner: Runner,
+    flow: Flow,
+    run: str,
+    files: dict[str, str],
+    entry: str,
+    map_spec: str,
+    where: str,
+    drift: bool,
+    logger: logging.Logger,
+) -> Outcome:
+    """plan (or, with `drift`, report drift) from the flow's snapshot."""
+    files, entry, derived = derive(runner, flow, run, files, entry, map_spec, where, logger)
+    inventory = runner.workspace(flow).write_input(run, files, entry)
+    common = {"input": files, "input_entry": entry, "input_sha256": input_sha256(files)}
+
+    if drift:
+        outcome = runner.drift(flow, run, inventory)
+        output = output_text(*derived, outcome.completed)
+        if outcome.error:
+            raise failed(outcome, output)
+        report = outcome.report
+        logger.info(f"drift: {report.counts()}")
+        return Outcome(
+            S.DRIFTED if report.has_drift else S.NO_CHANGES,
+            {"drift_report": report.doc, "summary": report.counts(), "output": output, **common},
+        )
+
+    outcome = runner.plan(flow, run, inventory)
+    output = output_text(*derived, outcome.completed)
+    if outcome.error:
+        raise failed(outcome, output)
+    plan = outcome.plan
+    logger.info(f"plan: {plan.summary()}")
+    fields = {
+        "plan": plan.raw.decode("utf-8"),
+        "plan_sha256": plan.sha256,
+        "summary": {**plan.summary(), "schema_preview": plan.schema_preview},
+        "output": output,
+        **common,
+    }
+    if plan.is_empty:
+        return Outcome(S.NO_CHANGES, fields)
+    return Outcome(S.AWAITING_APPROVAL, fields, finished=False)
+
+
+def approved_plan(plan_text: str, approved_sha256: str) -> Plan:
+    """the stored plan, refused unless it is the one that was approved."""
+    plan = Plan.from_bytes(plan_text.encode("utf-8"))
+    if not approved_sha256 or plan.sha256 != approved_sha256:
+        raise RunFailed("The stored plan is not the one that was approved.")
+    return plan
+
+
+def apply_run(
+    runner: Runner,
+    flow: Flow,
+    run: str,
+    files: dict[str, str],
+    entry: str,
+    plan: Plan,
+    approved_sha256: str,
+    resuming: bool,
+    before: str,
+    logger: logging.Logger,
+) -> Outcome:
+    """check the approved plan against the target, then apply it."""
+    inventory = runner.workspace(flow).write_input(run, files, entry)
+    outputs: list[Completed] = []
+    if not resuming:
+        # a resumed apply re-runs its plan on top of what already landed, so a
+        # re-plan would differ by construction. the journal carries it instead.
+        check = runner.check_stale(flow, run, inventory, plan)
+        outputs.append(check.completed)
+        if check.error:
+            raise failed(check, output_text(*outputs, before=before))
+        logger.info(f"stale check: the target {'changed' if check.stale else 'is unchanged'}")
+        if check.stale:
+            return Outcome(
+                S.STALE,
+                {
+                    "output": output_text(*outputs, before=before),
+                    "error": "The target changed since this plan was made; plan again.",
+                },
+            )
+
+    outcome = runner.apply(flow, run, plan, approved_sha256)
+    outputs.append(outcome.completed)
+    if outcome.error:
+        raise failed(outcome, output_text(*outputs, before=before))
+    logger.info(f"applied {outcome.report.applied_count} operations")
+    return Outcome(
+        S.APPLIED,
+        {
+            "apply_report": outcome.report.doc,
+            "output": output_text(*outputs, before=before),
+            "error": "",
+        },
+    )