path: root/webmap-download

#!/usr/bin/python3

from os import O_RDONLY, O_WRONLY, O_CREAT, O_TRUNC, O_CLOEXEC, O_PATH, O_DIRECTORY, O_TMPFILE
import os, sys
from fcntl import flock, LOCK_EX
import logging
from time import time, monotonic as time_monotonic
import argparse
import itertools
from pathlib import Path
from email.utils import parsedate_to_datetime, formatdate
from hashlib import sha1
from math import modf
import requests

import common

def download_trystream(url, **kwargs):
    max_tries = 10
    f = kwargs.pop('session', requests)
    for i in itertools.count(1):
        try:
            r = f.get(url, **kwargs, stream=True)
        except (requests.Timeout, requests.ConnectionError):
            if i < max_tries:
                logging.error('timeout')
                continue
            raise
        else:
            r.raise_for_status()
            return r

def download(url, dest, dir_fd=None, headers={}, max_size=None, session=requests):
    logging.info('Downloading %s…', url)
    destPath = Path(dest)
    dest_tmp = destPath.with_stem(f'.{destPath.stem}.new').as_posix()
    try:
        # delete any leftover
        os.unlink(dest_tmp, dir_fd=dir_fd)
    except FileNotFoundError:
        pass

    start = time_monotonic()
    r = download_trystream(url, headers=headers, session=session, timeout=30)
    if r.status_code == requests.codes.not_modified:
        # XXX shouldn't we call os.utime(dest) to bump its ctime here?
        # otherwise we'll make several queries and get multiple 304
        # replies if the file is used by multiple layers
        logging.info('%s: %d Not Modified', dest, r.status_code)
        return

    last_modified = r.headers.get('Last-Modified', None)
    if last_modified is not None:
        try:
            last_modified = parsedate_to_datetime(last_modified)
            last_modified = last_modified.timestamp()
        except ValueError:
            logging.exception('Could not parse Last-Modified value')
            last_modified = None

    # XXX we can't use TemporaryFile as it uses O_EXCL, cf.
    # https://discuss.python.org/t/temporaryfile-contextmanager-that-allows-creating-a-directory-entry-on-success/19094/2
    fd = os.open(os.path.dirname(dest), O_WRONLY|O_CLOEXEC|O_TMPFILE, mode=0o644, dir_fd=dir_fd)
    with os.fdopen(fd, mode='wb') as fp:
        size = 0
        for chunk in r.iter_content(chunk_size=2**16):
            size = size + len(chunk)
            if max_size is not None and size > max_size:
                raise Exception(f'Payload exceeds max-size ({max_size})')
            fp.write(chunk)
        r = None
        end = time_monotonic()

        # XXX unfortunately there is no way for linkat() to clobber the destination,
        # so we use a temporary file; it's racy, but thanks to O_TMPFILE better
        # (shorter race) than if we were dumping chunks in a named file descriptor
        os.link(f'/proc/self/fd/{fp.fileno()}', dest_tmp,
            dst_dir_fd=dir_fd, follow_symlinks=True)

    # no need to close fd here, it was taken care of by the context manager above

    try:
        if last_modified is not None:
            # XXX os.utime() doesn't work on file descriptors so we set mtime
            # after linkat() instead
            os.utime(dest_tmp, times=(last_modified, last_modified),
                dir_fd=dir_fd, follow_symlinks=False)
        os.rename(dest_tmp, dest, src_dir_fd=dir_fd, dst_dir_fd=dir_fd)

        elapsed = end - start
        logging.info("%s: Downloaded %s in %s (%s/s)", dest, format_bytes(size),
            format_time(elapsed), format_bytes(int(size/elapsed)))
    except Exception as e:
        try:
            os.unlink(dest_tmp, dir_fd=dir_fd)
        except Exception:
            pass
        raise e

def format_bytes(n):
    if n < 768:
        return f'{n}B'
    elif n < 786432:
        return f'{n/1024:.2f}kiB'
    elif n < 805306368:
        return f'{n/1048576:.2f}MiB'
    else:
        return f'{n/1073741824:.2f}GiB'

def format_time(s):
    fs, s = modf(s)
    m, s = divmod(int(s), 60)
    h, m = divmod(m, 60)
    return f'{h:02d}:{m:02d}:{s + fs:06.3f}'

if __name__ == '__main__':
    logging.basicConfig(format='%(levelname)s: %(message)s', level=logging.INFO)

    parser = argparse.ArgumentParser(description='Download or update GIS layers.')
    parser.add_argument('--cachedir', default=None,
        help='destination directory for downloaded files (default: .)')
    parser.add_argument('--lockdir', default=None,
        help='directory for lock files (default: value of --cachedir option)')
    parser.add_argument('--debug', action='store_true', help=argparse.SUPPRESS)
    parser.add_argument('--exit-code', action=argparse.BooleanOptionalAction,
        help='whether to exit with status 1 in case of download failures')
    parser.add_argument('groupname', nargs='*', help='Group(s) to process')
    args = parser.parse_args()

    if args.debug:
        from http.client import HTTPConnection
        HTTPConnection.debuglevel = 1
        logging.getLogger().setLevel(logging.DEBUG)
        requests_log = logging.getLogger("urllib3")
        requests_log.setLevel(logging.DEBUG)
        requests_log.propagate = True

    common.load_config(groupnames=None if args.groupname == [] else args.groupname)

    sources = []
    for name, layerdefs in common.config.get('layers', {}).items():
        for layerdef in layerdefs:
            sourcedef = layerdef.get('source', {})
            sourcedef['layername'] = name
            sources.append(sourcedef)

    # intentionally leave the dirfd open until the program terminates
    opendir_args = O_RDONLY|O_CLOEXEC|O_PATH|O_DIRECTORY
    destdir_fd = None       if args.cachedir is None else os.open(args.cachedir, opendir_args)
    lockdir_fd = destdir_fd if args.lockdir  is None else os.open(args.lockdir, opendir_args)

    if ((destdir_fd is None and lockdir_fd is None) or
            (destdir_fd is not None and lockdir_fd is not None and destdir_fd == lockdir_fd)):
        lockdir_is_destdir = True
    else:
        destdir_st = os.fstat(destdir_fd)
        lockdir_st = os.fstat(lockdir_fd)
        lockdir_is_destdir = destdir_st.st_ino == lockdir_st.st_ino

    sessionRequests = requests.Session()

    rv = 0
    downloads = set()
    for source in sources:
        dl = source.get('download', None)
        script = None if dl is None else dl.get('script', None)
        if script is not None and script != os.path.basename(__file__):
            logging.info('Layer "%s" is not for us (%s != %s), skipping',
                source['layername'],
                script, os.path.basename(__file__))
            continue

        url = None if dl is None else dl.get('url', None)
        if url is None:
            logging.error('Layer "%s" has no source URL, ignoring',
                source['layername'])
            continue

        cache = source.get('cache', None)
        dest = None if cache is None else cache.get('path', None)
        if dest is None:
            raise Exception('Impossible')
        elif url in downloads:
            logging.info('%s was already downloaded, skipping', dest)
            continue

        headers = {}
        user_agent = common.config.get('User-Agent', None)
        if user_agent is not None:
            headers['User-Agent'] = user_agent
        dest = str(dest) # convert from Path()

        try:
            # create parent directories
            destdir = os.path.dirname(dest)
            common.makedirs(destdir, mode=0o755, dir_fd=destdir_fd, exist_ok=True, logging=logging)

            # place an exclusive lock on a lockfile as the destination can be used by other layers
            # hence might be updated in parallel
            if lockdir_is_destdir:
                lockfile = dest + '.lck'
            else:
                # use a flat hierarchy when lockdir != destdir as this avoids leaving empty directories
                # behind when removing left overs with tmpfiles.d(5)
                lockfile = sha1(dest.encode('utf-8')).hexdigest() + '.lck'
            # use O_TRUNC to bump lockfile's mtime
            lock_fd = os.open(lockfile, O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, mode=0o644, dir_fd=lockdir_fd)
            try:
                logging.debug('flock("%s", LOCK_EX)', lockfile)
                flock(lock_fd, LOCK_EX)
                try:
                    st = os.stat(dest, dir_fd=destdir_fd)
                except (OSError, ValueError):
                    # the file doesn't exist, or stat() failed for some reason
                    pass
                else:
                    max_age = cache.get('max-age', 6*3600) # 6h
                    if max_age is not None:
                        s = max_age + max(st.st_ctime, st.st_mtime) - time()
                        if s > 0:
                            logging.info('%s: Too young, try again in %s',
                                dest, format_time(s))
                            continue
                    headers['If-Modified-Since'] = formatdate(timeval=st.st_mtime, localtime=False, usegmt=True)
                max_size = dl.get('max-size', 2**26) # 64MiB
                download(url, dest, dir_fd=destdir_fd, max_size=max_size,
                    headers=headers, session=sessionRequests)
                downloads.add(url)
            finally:
                os.close(lock_fd)
        except Exception:
            logging.exception('Could not download %s as %s', url, dest)
            if args.exit_code:
                rv = 1
    exit(rv)