示例#1
0
    def process(self, dirName, nzbName=None, status=0, clientAgent="manual", download_id="", inputCategory=None):
        # auto-detect correct section
        section = nzbtomedia.CFG.findsection(inputCategory)
        if not section:
            logger.error(
                "We were unable to find a section for category %s, please check your autoProcessMedia.cfg file." % inputCategory)
            return 1

        status = int(status)

        host = nzbtomedia.CFG[section][inputCategory]["host"]
        port = nzbtomedia.CFG[section][inputCategory]["port"]
        apikey = nzbtomedia.CFG[section][inputCategory]["apikey"]
        method = nzbtomedia.CFG[section][inputCategory]["method"]
        delete_failed = int(nzbtomedia.CFG[section][inputCategory]["delete_failed"])
        wait_for = int(nzbtomedia.CFG[section][inputCategory]["wait_for"])

        try:
            ssl = int(nzbtomedia.CFG[section][inputCategory]["ssl"])
        except:
            ssl = 0
        try:
            web_root = nzbtomedia.CFG[section][inputCategory]["web_root"]
        except:
            web_root = ""
        try:
            remote_path = nzbtomedia.CFG[section][inputCategory]["remote_path"]
        except:
            remote_path = None

        if ssl:
            protocol = "https://"
        else:
            protocol = "http://"

        baseURL = "%s%s:%s%s/api/%s" % (protocol, host, port, web_root, apikey)

        imdbid = find_imdbid(dirName, nzbName)
        release = self.get_release(baseURL, imdbid, download_id)

        # pull info from release found if available
        release_id = None
        media_id = None
        downloader = None
        release_status_old = None
        if len(release) == 1:
            try:
                release_id = release.keys()[0]
                media_id = release[release_id]['media_id']
                download_id = release[release_id]['download_info']['id']
                downloader = release[release_id]['download_info']['downloader']
                release_status_old = release[release_id]['status']
            except:
                pass

        process_all_exceptions(nzbName.lower(), dirName)
        nzbName, dirName = convert_to_ascii(nzbName, dirName)

        if status == 0:
            if nzbtomedia.TRANSCODE == 1:
                result = Transcoder().Transcode_directory(dirName)
                if result == 0:
                    logger.debug("Transcoding succeeded for files in %s" % (dirName), section)
                else:
                    logger.warning("Transcoding failed for files in %s" % (dirName), section)

            if method == "manage":
                command = "/manage.update"
            else:
                command = "/renamer.scan"

            params = {}
            if download_id:
                params['downloader'] = downloader or clientAgent
                params['download_id'] = download_id

            params['media_folder'] = dirName
            if remote_path:
                dirName_new = os.path.join(remote_path, os.path.basename(dirName)).replace("\\", "/")
                params['media_folder'] = dirName_new

            url = "%s%s" % (baseURL, command)

            logger.debug("Opening URL: %s" % (url), section)

            logger.postprocess("Starting %s scan for %s" % (method, nzbName), section)

            try:
                r = requests.get(url, params=params)
            except requests.ConnectionError:
                logger.error("Unable to open URL", section)
                return 1  # failure

            result = r.json()
            if result['success']:
                logger.postprocess("SUCCESS: Finished %s scan for folder %s" % (method, dirName), section)
            else:
                logger.error("FAILED: %s scan was unable to finish for folder %s. exiting!" % (method, dirName),
                             section)
                return 1  # failure

        else:
            logger.postprocess("FAILED DOWNLOAD DETECTED FOR %s" % (nzbName), section)

            if delete_failed and os.path.isdir(dirName) and not os.path.dirname(dirName) == dirName:
                logger.postprocess("Deleting failed files and folder %s" % dirName, section)
                delete(dirName)

            if not download_id:
                logger.error("Could not find a downloaded movie in the database matching %s, exiting!" % nzbName,
                             section)
                return 1  # failure

            logger.postprocess("Setting failed release %s to ignored ..." % (nzbName), section)

            url = baseURL + "/release.ignore"
            logger.debug("Opening URL: %s" % (url), section)

            try:
                r = requests.get(url, params={'id': release_id})
            except requests.ConnectionError:
                logger.error("Unable to open URL %s" % (url), section)
                return 1  # failure

            result = r.json()
            if result['success']:
                logger.postprocess("SUCCESS: %s has been set to ignored ..." % (nzbName), section)
            else:
                logger.warning("FAILED: Unable to set %s to ignored!" % (nzbName), section)

            logger.postprocess("Trying to snatch the next highest ranked release.", section)

            url = "%s/movie.searcher.try_next" % (baseURL)
            logger.debug("Opening URL: %s" % (url), section)

            try:
                r = requests.get(url, params={'media_id': media_id})
            except requests.ConnectionError:
                logger.error("Unable to open URL %s" % (url), section)
                return 1  # failure

            result = r.json()
            if result['success']:
                logger.postprocess("SUCCESS: Snatched the next highest release ...", section)
                return 0
            else:
                logger.postprocess("FAILED: Unable to find a higher ranked release then %s to snatch!" % (nzbName),
                                   section)
                return 1

        # we will now check to see if CPS has finished renaming before returning to TorrentToMedia and unpausing.
        timeout = time.time() + 60 * wait_for
        while (time.time() < timeout):  # only wait 2 (default) minutes, then return.
            logger.postprocess("Checking for status change, please stand by ...", section)
            release = self.get_release(baseURL, imdbid, download_id, release_id)
            if release:
                try:
                    release_status_new = release[release_id]['status']
                    if release_status_new != release_status_old:
                        logger.postprocess("SUCCESS: Release %s has now been marked with a status of [%s]" % (
                            nzbName, str(release_status_new).upper()), section)
                        return 0  # success
                except:
                    pass

            # pause and let CouchPotatoServer catch its breath
            time.sleep(10 * wait_for)

        # The status hasn't changed. we have waited 2 minutes which is more than enough. uTorrent can resule seeding now.
        logger.warning(
            "%s does not appear to have changed status after %s minutes, Please check your logs." % (nzbName, wait_for),
            section)
        return 1  # failure
    def processEpisode(self, dirName, nzbName=None, failed=False, clientAgent = "manual", inputCategory=None):
        # auto-detect correct section
        section = nzbtomedia.CFG.findsection(inputCategory)
        if not section:
            logger.error(
                "We were unable to find a section for category %s, please check your autoProcessMedia.cfg file." % inputCategory)
            return 1

        # auto-detect correct fork
        fork, fork_params = autoFork(inputCategory)

        status = int(failed)

        host = nzbtomedia.CFG[section][inputCategory]["host"]
        port = nzbtomedia.CFG[section][inputCategory]["port"]
        username = nzbtomedia.CFG[section][inputCategory]["username"]
        password = nzbtomedia.CFG[section][inputCategory]["password"]

        try:
            apikey = nzbtomedia.CFG[section][inputCategory]["apikey"]
        except:
            apikey = ""
        try:
            ssl = int(nzbtomedia.CFG[section][inputCategory]["ssl"])
        except:
            ssl = 0
        try:
            web_root = nzbtomedia.CFG[section][inputCategory]["web_root"]
        except:
            web_root = ""
        try:
            delete_failed = int(nzbtomedia.CFG[section][inputCategory]["delete_failed"])
        except:
            delete_failed = 0
        try:
            nzbExtractionBy = nzbtomedia.CFG[section][inputCategory]["nzbExtractionBy"]
        except:
            nzbExtractionBy = "Downloader"
        try:
            process_method = nzbtomedia.CFG[section][inputCategory]["process_method"]
        except:
            process_method = None
        try:
            Torrent_NoLink = int(nzbtomedia.CFG[section][inputCategory]["Torrent_NoLink"])
        except:
            Torrent_NoLink = 0
        try:
            remote_path = nzbtomedia.CFG[section][inputCategory]["remote_path"]
        except:
            remote_path = None

        if not os.path.isdir(dirName) and os.path.isfile(dirName): # If the input directory is a file, assume single file download and split dir/name.
            dirName = os.path.split(os.path.normpath(dirName))[0]

        SpecificPath = os.path.join(dirName, str(nzbName))
        cleanName = os.path.splitext(SpecificPath)
        if cleanName[1] == ".nzb":
            SpecificPath = cleanName[0]
        if os.path.isdir(SpecificPath):
            dirName = SpecificPath

        if fork not in nzbtomedia.SICKBEARD_TORRENT or (clientAgent in ['nzbget','sabnzbd'] and nzbExtractionBy != "Destination"):
            if nzbName:
                process_all_exceptions(nzbName.lower(), dirName)
                nzbName, dirName = convert_to_ascii(nzbName, dirName)

            # Now check if tv files exist in destination. Eventually extraction may be done here if nzbExtractionBy == TorrentToMedia
            video = int(0)
            for dirpath, dirnames, filenames in os.walk(dirName):
                for file in filenames:
                    filePath = os.path.join(dirpath, file)
                    fileExtension = os.path.splitext(file)[1]
                    if fileExtension in nzbtomedia.MEDIACONTAINER:  # If the file is a video file
                        if is_sample(filePath, nzbName, nzbtomedia.MINSAMPLESIZE, nzbtomedia.SAMPLEIDS):
                            logger.debug("Removing sample file: %s" % (filePath), section)
                            os.unlink(filePath)  # remove samples
                        else:
                            video = video + 1
            if video > 0:  # Check that a video exists. if not, assume failed.
                flatten(dirName) # to make sure SickBeard can find the video (not in sub-folder)
            elif clientAgent == "manual":
                logger.warning("No media files found in directory %s to manually process." % (dirName), section)
                return 0  # Success (as far as this script is concerned)
            else:
                logger.warning("No media files found in directory %s. Processing this as a failed download" % (dirName), section)
                status = int(1)
                failed = True

        # configure SB params to pass
        fork_params['quiet'] = 1
        if nzbName is not None:
            fork_params['nzbName'] = nzbName

        for param in copy.copy(fork_params):
            if param == "failed":
                fork_params[param] = failed

            if param in ["dirName", "dir"]:
                fork_params[param] = dirName
                if remote_path:
                    dirName_new = os.path.join(remote_path, os.path.basename(dirName)).replace("\\", "/")
                    fork_params[param] = dirName_new

            if param == "process_method":
                if fork in nzbtomedia.SICKBEARD_TORRENT and Torrent_NoLink == 1 and not clientAgent in ['nzbget','sabnzbd']: #use default SickBeard settings here.
                    del fork_params[param]
                if process_method:
                    fork_params[param] = process_method
                else:
                    del fork_params[param]

        # delete any unused params so we don't pass them to SB by mistake
        [fork_params.pop(k) for k,v in fork_params.items() if v is None]

        if status == 0:
            logger.postprocess("SUCCESS: The download succeeded, sending a post-process request", section)
        else:
            if fork in nzbtomedia.SICKBEARD_FAILED:
                logger.postprocess("FAILED: The download failed. Sending 'failed' process request to %s branch" % (fork), section)
            else:
                logger.postprocess("FAILED: The download failed. %s branch does not handle failed downloads. Nothing to process" % (fork), section)
                if delete_failed and os.path.isdir(dirName) and not os.path.dirname(dirName) == dirName:
                    logger.postprocess("Deleting failed files and folder %s" % (dirName), section)
                    delete(dirName)
                return 0 # Success (as far as this script is concerned)

        if status == 0 and nzbtomedia.TRANSCODE == 1: # only transcode successful downlaods
            result = Transcoder().Transcode_directory(dirName)
            if result == 0:
                logger.debug("SUCCESS: Transcoding succeeded for files in %s" % (dirName), section)
            else:
                logger.warning("FAILED: Transcoding failed for files in %s" % (dirName), section)

        if ssl:
            protocol = "https://"
        else:
            protocol = "http://"

        url = None
        if section == "SickBeard":
            url = "%s%s:%s%s/home/postprocess/processEpisode" % (protocol,host,port,web_root)
        elif section == "NzbDrone":
            url = "%s%s:%s%s/api/command" % (protocol, host, port, web_root)

        logger.debug("Opening URL: %s" % (url),section)

        try:
            r = None
            if section == "SickBeard":
                r = requests.get(url, auth=(username, password), params=fork_params, stream=True)
            elif section == "NzbDrone":
                params = {"name": "DownloadedEpisodesScan", "path": dirName}
                headers = {"X-Api-Key": apikey}
                r = requests.get(url, params=params, headers=headers, stream=True)
        except requests.ConnectionError:
            logger.error("Unable to open URL: %s" % (url), section)
            return 1 # failure

        for line in r.iter_lines():
            if line: logger.postprocess("%s" % (line), section)

        if status != 0 and delete_failed and not os.path.dirname(dirName) == dirName:
            logger.postprocess("Deleting failed files and folder %s" % (dirName),section)
            delete(dirName)
        return 0 # Success
    def process(self,
                dirName,
                nzbName=None,
                status=0,
                clientAgent="manual",
                download_id="",
                inputCategory=None):
        # auto-detect correct section
        section = nzbtomedia.CFG.findsection(inputCategory)
        if not section:
            logger.error(
                "We were unable to find a section for category %s, please check your autoProcessMedia.cfg file."
                % inputCategory)
            return 1

        status = int(status)

        host = nzbtomedia.CFG[section][inputCategory]["host"]
        port = nzbtomedia.CFG[section][inputCategory]["port"]
        apikey = nzbtomedia.CFG[section][inputCategory]["apikey"]
        method = nzbtomedia.CFG[section][inputCategory]["method"]
        delete_failed = int(
            nzbtomedia.CFG[section][inputCategory]["delete_failed"])
        wait_for = int(nzbtomedia.CFG[section][inputCategory]["wait_for"])

        try:
            ssl = int(nzbtomedia.CFG[section][inputCategory]["ssl"])
        except:
            ssl = 0
        try:
            web_root = nzbtomedia.CFG[section][inputCategory]["web_root"]
        except:
            web_root = ""
        try:
            remote_path = nzbtomedia.CFG[section][inputCategory]["remote_path"]
        except:
            remote_path = None

        if ssl:
            protocol = "https://"
        else:
            protocol = "http://"

        baseURL = "%s%s:%s%s/api/%s" % (protocol, host, port, web_root, apikey)

        imdbid = find_imdbid(dirName, nzbName)
        release = self.get_release(baseURL, imdbid, download_id)

        # pull info from release found if available
        release_id = None
        media_id = None
        downloader = None
        release_status_old = None
        if len(release) == 1:
            try:
                release_id = release.keys()[0]
                media_id = release[release_id]['media_id']
                download_id = release[release_id]['download_info']['id']
                downloader = release[release_id]['download_info']['downloader']
                release_status_old = release[release_id]['status']
            except:
                pass

        process_all_exceptions(nzbName.lower(), dirName)
        nzbName, dirName = convert_to_ascii(nzbName, dirName)

        if status == 0:
            if nzbtomedia.TRANSCODE == 1:
                result = Transcoder().Transcode_directory(dirName)
                if result == 0:
                    logger.debug(
                        "Transcoding succeeded for files in %s" % (dirName),
                        section)
                else:
                    logger.warning(
                        "Transcoding failed for files in %s" % (dirName),
                        section)

            if method == "manage":
                command = "/manage.update"
            else:
                command = "/renamer.scan"

            params = {}
            if download_id:
                params['downloader'] = downloader or clientAgent
                params['download_id'] = download_id

            params['media_folder'] = dirName
            if remote_path:
                dirName_new = os.path.join(remote_path,
                                           os.path.basename(dirName)).replace(
                                               "\\", "/")
                params['media_folder'] = dirName_new

            url = "%s%s" % (baseURL, command)

            logger.debug("Opening URL: %s" % (url), section)

            logger.postprocess("Starting %s scan for %s" % (method, nzbName),
                               section)

            try:
                r = requests.get(url, params=params)
            except requests.ConnectionError:
                logger.error("Unable to open URL", section)
                return 1  # failure

            result = r.json()
            if result['success']:
                logger.postprocess(
                    "SUCCESS: Finished %s scan for folder %s" %
                    (method, dirName), section)
            else:
                logger.error(
                    "FAILED: %s scan was unable to finish for folder %s. exiting!"
                    % (method, dirName), section)
                return 1  # failure

        else:
            logger.postprocess("FAILED DOWNLOAD DETECTED FOR %s" % (nzbName),
                               section)

            if delete_failed and os.path.isdir(
                    dirName) and not os.path.dirname(dirName) == dirName:
                logger.postprocess(
                    "Deleting failed files and folder %s" % dirName, section)
                delete(dirName)

            if not download_id:
                logger.error(
                    "Could not find a downloaded movie in the database matching %s, exiting!"
                    % nzbName, section)
                return 1  # failure

            logger.postprocess(
                "Setting failed release %s to ignored ..." % (nzbName),
                section)

            url = baseURL + "/release.ignore"
            logger.debug("Opening URL: %s" % (url), section)

            try:
                r = requests.get(url, params={'id': release_id})
            except requests.ConnectionError:
                logger.error("Unable to open URL %s" % (url), section)
                return 1  # failure

            result = r.json()
            if result['success']:
                logger.postprocess(
                    "SUCCESS: %s has been set to ignored ..." % (nzbName),
                    section)
            else:
                logger.warning(
                    "FAILED: Unable to set %s to ignored!" % (nzbName),
                    section)

            logger.postprocess(
                "Trying to snatch the next highest ranked release.", section)

            url = "%s/movie.searcher.try_next" % (baseURL)
            logger.debug("Opening URL: %s" % (url), section)

            try:
                r = requests.get(url, params={'media_id': media_id})
            except requests.ConnectionError:
                logger.error("Unable to open URL %s" % (url), section)
                return 1  # failure

            result = r.json()
            if result['success']:
                logger.postprocess(
                    "SUCCESS: Snatched the next highest release ...", section)
                return 0
            else:
                logger.postprocess(
                    "FAILED: Unable to find a higher ranked release then %s to snatch!"
                    % (nzbName), section)
                return 1

        # we will now check to see if CPS has finished renaming before returning to TorrentToMedia and unpausing.
        timeout = time.time() + 60 * wait_for
        while (time.time() <
               timeout):  # only wait 2 (default) minutes, then return.
            logger.postprocess(
                "Checking for status change, please stand by ...", section)
            release = self.get_release(baseURL, imdbid, download_id,
                                       release_id)
            if release:
                try:
                    release_status_new = release[release_id]['status']
                    if release_status_new != release_status_old:
                        logger.postprocess(
                            "SUCCESS: Release %s has now been marked with a status of [%s]"
                            % (nzbName, str(release_status_new).upper()),
                            section)
                        return 0  # success
                except:
                    pass

            # pause and let CouchPotatoServer catch its breath
            time.sleep(10 * wait_for)

        # The status hasn't changed. we have waited 2 minutes which is more than enough. uTorrent can resule seeding now.
        logger.warning(
            "%s does not appear to have changed status after %s minutes, Please check your logs."
            % (nzbName, wait_for), section)
        return 1  # failure