Source code for latch_cli.services.local_dev_old

import asyncio
import json
import os
import signal
import subprocess
import sys
import termios
import time
import traceback
import tty
from enum import Enum
from http.client import HTTPException
from pathlib import Path
from textwrap import dedent
from typing import Callable, Optional, Tuple, Union

import aioconsole
import asyncssh
import boto3
import click
import websockets.client as client
import websockets.exceptions
from latch_sdk_config.latch import config
from watchfiles import awatch

from latch.utils import current_workspace, retrieve_or_login
from latch_cli.constants import docker_image_name_illegal_pat
from latch_cli.tinyrequests import post
from latch_cli.utils import TemporarySSHCredentials, identifier_suffix_from_str


def _get_workflow_name(pkg_root: Path, snakemake: bool) -> str:
    if snakemake:
        import latch.types.metadata as metadata
        from latch_cli.snakemake.utils import load_snakemake_metadata

        load_snakemake_metadata(pkg_root)
        if metadata._snakemake_metadata is None:
            click.secho(f"Unable to find latch_metadata in {pkg_root}.", fg="red")
            raise click.exceptions.Exit(1)
        return f"{metadata._snakemake_metadata.name}_jit_register"
    else:
        from flytekit.core.context_manager import FlyteEntities
        from flytekit.core.workflow import PythonFunctionWorkflow

        from latch_cli.centromere.utils import _import_flyte_objects

        _import_flyte_objects([pkg_root])
        for entity in FlyteEntities.entities:
            if isinstance(entity, PythonFunctionWorkflow):
                return entity.name

        click.secho(
            dedent(f"""
            Unable to find a workflow in {pkg_root}: Check that

            1. {pkg_root} is a valid workflow directory, and
            2. Your workflow function has the @workflow decorator.
            3. Pass in the --snakemake flag if you are using snakemake.
            """).strip("\n"),
            fg="red",
        )
        raise click.exceptions.Exit(1)


def _get_latest_image(pkg_root: Path, snakemake: bool) -> str:
    click.secho("Image name: ", fg="blue", nl=False)

    ws_id = current_workspace()
    if int(ws_id) < 10:
        ws_id = f"x{ws_id}"

    wf_name = identifier_suffix_from_str(
        _get_workflow_name(pkg_root, snakemake)
    ).lower()
    wf_name = docker_image_name_illegal_pat.sub("_", wf_name)

    registry_name = f"{ws_id}_{wf_name}"

    click.secho(registry_name, italic=True, bold=True)

    resp = post(
        config.api.workflow.get_latest,
        headers={"Authorization": f"Bearer {retrieve_or_login()}"},
        json={
            "registry_name": registry_name,
            "ws_account_id": current_workspace(),
        },
    )

    try:
        resp.raise_for_status()
        click.secho("Found image. Using version ", nl=False)
        latest_version = resp.json()["version"]
        click.secho(latest_version, italic=True, bold=True)
    except HTTPException:
        click.secho("Could not find any images. Retrying with new image name.\n")
        click.secho("Image name: ", fg="blue", nl=False)

        registry_name = f"{ws_id}_{pkg_root.name}"

        click.secho(registry_name, italic=True, bold=True)

        resp = post(
            config.api.workflow.get_latest,
            headers={"Authorization": f"Bearer {retrieve_or_login()}"},
            json={
                "registry_name": registry_name,
                "ws_account_id": current_workspace(),
            },
        )

        try:
            resp.raise_for_status()
            click.secho("Found image. Using version ", nl=False)
            latest_version = resp.json()["version"]
            click.secho(latest_version, italic=True, bold=True)
        except HTTPException as e:
            click.secho("Could not find any images.\n")

            click.secho(
                dedent("""
                There was an issue getting your workflow's docker image.

                Please make sure you've registered your workflow at least once.
                """).strip("\n"),
                fg="red",
            )
            raise click.exceptions.Exit(1)

    return f"{config.dkr_repo}/{registry_name}:{latest_version}"


rsync_stop_event = asyncio.Event()


async def _watch_and_rsync(
    pkg_root: Path,
    ip: str,
    ssh_port: str,
    key_path: Path,
    startup_grace_period: int = 10,
):
    async def _rsync() -> Tuple[int, str, str]:
        # todo(aidan): set up host key rather than disabling strict host checking
        command = [
            "rsync",
            f"--rsh=ssh -p {ssh_port} -i {str(key_path)} -o StrictHostKeyChecking=no",
            "--compress",
            "--recursive",
            "--links",
            "--times",
            "--devices",
            "--specials",
            f"{str(pkg_root)}/",
            f"root@{ip}:/root/",
        ]
        process = await asyncio.subprocess.create_subprocess_exec(
            *command,
            stdout=asyncio.subprocess.PIPE,
            stderr=asyncio.subprocess.PIPE,
        )
        rval = await process.wait()
        stdout, stderr = await process.communicate()
        return rval, stdout.decode("utf-8"), stderr.decode("utf-8")

    # run rsync a maximum of once every 500ms
    watcher = awatch(str(pkg_root), debounce=500, step=500, stop_event=rsync_stop_event)

    start_time = time.time()
    while time.time() - start_time < startup_grace_period:
        returncode, _, _ = await _rsync()
        if returncode == 0:
            break

        await asyncio.sleep(1)

    # todo(aidan): upgrade watchfiles and remove this error suppression after the next release of watchfiles
    # https://github.com/samuelcolvin/watchfiles/issues/200
    try:
        async for _ in watcher:
            rval, _, stderr = await _rsync()
            if rval != 0:
                await aioconsole.aprint(
                    f"Error running rsync: {stderr}", use_stderr=True
                )
    except RuntimeError as e:
        if not str(e) == "Already borrowed":
            raise e


class _MessageType(Enum):
    resize = "resize"
    body = "body"
    exit = "exit"


async def _get_messages(
    ws: client.WebSocketClientProtocol,
    show_output: bool,
):
    async for message in ws:
        msg = json.loads(message)
        if msg.get("Type") == _MessageType.exit.value:
            return
        if show_output:
            await aioconsole.aprint(
                msg.get("Body"),
                end="",
                flush=True,
            )


async def _send_message(
    ws: client.WebSocketClientProtocol,
    message: Union[str, bytes],
    typ: Optional[_MessageType] = None,
):
    if isinstance(message, bytes):
        message = message.decode("utf-8")
    if typ is None:
        typ = _MessageType.body
    body = {"Body": message, "Type": typ.value}
    await ws.send(json.dumps(body))
    # yield control back to the event loop,
    # see https://github.com/aaugustin/websockets/issues/867
    await asyncio.sleep(0)


async def _send_resize_message(
    ws: client.WebSocketClientProtocol,
    term_width: int,
    term_height: int,
):
    await _send_message(
        ws,
        json.dumps(
            {
                "Width": term_width,
                "Height": term_height,
            }
        ),
        typ=_MessageType.resize,
    )


async def _shell_session(
    ws: client.WebSocketClientProtocol,
):
    old_settings_stdin = termios.tcgetattr(sys.stdin.fileno())
    tty.setraw(sys.stdin)

    loop = asyncio.get_event_loop()
    reader = asyncio.StreamReader(loop=loop)

    stdin2 = os.dup(sys.stdin.fileno())
    stdout2 = os.dup(sys.stdout.fileno())

    old_sigwinch_handler = signal.getsignal(signal.SIGWINCH)

    await loop.connect_read_pipe(
        lambda: asyncio.StreamReaderProtocol(
            reader,
            loop=loop,
        ),
        os.fdopen(stdin2),
    )

    writer_transport, writer_protocol = await loop.connect_write_pipe(
        lambda: asyncio.streams.FlowControlMixin(loop=loop),
        os.fdopen(stdout2),
    )
    writer = asyncio.streams.StreamWriter(
        writer_transport,
        writer_protocol,
        None,
        loop,
    )

    resize_event_queue = asyncio.Queue()

    def new_sigwinch_handler(signum, frame):
        if isinstance(old_sigwinch_handler, Callable):
            old_sigwinch_handler(signum, frame)

        term_width, term_height = os.get_terminal_size()
        resize_event_queue.put_nowait((term_width, term_height))

    signal.signal(signal.SIGWINCH, new_sigwinch_handler)

    # get initial terminal size and send it over
    term_width, term_height = os.get_terminal_size()
    resize_event_queue.put_nowait((term_width, term_height))

    async def resize_task():
        while True:
            (term_width, term_height) = await resize_event_queue.get()
            await _send_resize_message(ws, term_width, term_height)

    async def input_task():
        while True:
            message = await reader.read(1)
            await _send_message(ws, message)

    async def output_task():
        async for output in ws:
            obj: dict = json.loads(output)
            if obj.get("Type") == "exit":
                io_task.cancel()
                return

            message = obj.get("Body")
            if message is not None:
                if isinstance(message, str):
                    message = message.encode("utf-8")
                writer.write(message)
            await writer.drain()
            await asyncio.sleep(0)

    try:
        io_task = asyncio.gather(input_task(), output_task(), resize_task())
        await io_task
    except asyncio.CancelledError: ...
    finally:
        termios.tcsetattr(sys.stdin.fileno(), termios.TCSANOW, old_settings_stdin)
        signal.signal(signal.SIGWINCH, old_sigwinch_handler)


async def _run_local_dev_session(pkg_root: Path, snakemake: bool):
    # hit the endpoint to make sure that a workflow image exists in ecr before
    # doing anything
    image_name_tagged = _get_latest_image(pkg_root, snakemake)
    key_path = pkg_root / ".latch" / "ssh_key"

    with TemporarySSHCredentials(key_path) as ssh:
        headers = {"Authorization": f"Bearer {retrieve_or_login()}"}

        try:
            cent_resp = post(
                config.api.centromere.provision,
                headers=headers,
                json={"public_key": ssh.public_key},
            )
            cent_resp.raise_for_status()

            init_local_dev_resp = post(
                config.api.centromere.start_local_dev,
                headers=headers,
                json={
                    "ws_account_id": current_workspace(),
                    "pkg_root": pkg_root.name,
                },
            )
            init_local_dev_resp.raise_for_status()
        except Exception as e:
            raise ValueError(f"Unable to provision a remote instance: {e}")

        if init_local_dev_resp.status_code == 403:
            raise ValueError("You are not authorized to use this feature.")

        try:
            cent_data = cent_resp.json()
            centromere_ip = cent_data["ip"]
        except KeyError as e:
            raise ValueError(f"Malformed response from provision endpoint: missing {e}")

        try:
            init_local_dev_data = init_local_dev_resp.json()
            docker_info = init_local_dev_data["docker"]
            access_key = docker_info["tmp_access_key"]
            secret_key = docker_info["tmp_secret_key"]
            session_token = docker_info["tmp_session_token"]
            port = init_local_dev_data["port"]
            ssh_port = init_local_dev_data["ssh_port"]
        except KeyError as e:
            raise ValueError(
                f"Malformed response from initialization endpoint: missing {e}"
            )

        # ecr setup
        ws_id = current_workspace()
        if int(ws_id) < 10:
            ws_id = f"x{ws_id}"

        try:
            ecr_client = boto3.session.Session(
                aws_access_key_id=access_key,
                aws_secret_access_key=secret_key,
                aws_session_token=session_token,
                region_name="us-west-2",
            ).client("ecr")

            ecr_auth = ecr_client.get_authorization_token()
            docker_access_token = ecr_auth["authorizationData"][0]["authorizationToken"]

        except Exception as err:
            raise ValueError(f"unable to retrieve an ecr login token") from err

        # todo(aidan): edit known hosts file with centromere ip and address to allow strict host checking
        try:
            async for ws in client.connect(
                f"ws://{centromere_ip}:{port}/ws",
                close_timeout=0,
                extra_headers=headers,
            ):
                try:
                    await aioconsole.aprint(
                        "Successfully connected to remote instance."
                    )
                    try:
                        await aioconsole.aprint("Setting up local sync...")
                        watch_task = asyncio.create_task(
                            _watch_and_rsync(
                                pkg_root, centromere_ip, ssh_port, key_path
                            )
                        )
                        await aioconsole.aprint("Done.")

                        await _send_message(ws, docker_access_token)
                        await _send_message(ws, image_name_tagged)
                        await aioconsole.aprint(f"Pulling {image_name_tagged}... ")

                        await _get_messages(ws, show_output=False)
                        await aioconsole.aprint("Image successfully pulled.\n")

                        await _get_messages(ws, show_output=True)

                        await _shell_session(ws)
                        rsync_stop_event.set()
                        await watch_task

                    except websockets.exceptions.ConnectionClosed:
                        continue
                except Exception as e:
                    await aioconsole.aprint(traceback.format_exc())
                finally:
                    await aioconsole.aprint("Exiting local development session")
                    await ws.close()
                    close_resp = post(
                        config.api.centromere.stop_local_dev,
                        headers={"Authorization": f"Bearer {retrieve_or_login()}"},
                    )
                    close_resp.raise_for_status()
                    return
        except asyncssh.ProtocolError as e:
            if "Too many authentication failures" in str(e):
                click.secho(
                    dedent("""
                    Too many authentication failures. Try resetting your ssh-agent with

                        $ ssh-add -D

                    and trying again."""),
                    fg="red",
                )
            else:
                raise e


[docs]def local_development(pkg_root: Path, snakemake: bool): """Starts a REPL that allows a user to interactively run tasks to help with debugging during workflow development. In this REPL, you can run tasks or scripts and make edits to them without having to reregister your workflow. You can also get a shell into a container with the same environment as the one that the workflow runs in, to help debug installation issues. See the full documentation for `Local Development` for more info. Like `get_executions`, this should only be called from the CLI for best results. Args: pkg_root: A path that points to a valid workflow directory (see the docs for `register`) """ # ensure that rsync is installed try: subprocess.run( ["rsync", "--version"], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, check=True, ) except FileNotFoundError: raise ValueError(dedent(""" rsync is required for latch develop. Please install rsync and try again linux: apt install rsync mac: brew install rsync """)) asyncio.run(_run_local_dev_session(pkg_root, snakemake))