About

Microservice to manage the downloads of biomaj.

A protobuf interface is available in biomaj_download/message/message_pb2.py to exchange messages between BioMAJ and the download service. Messages go through RabbitMQ (to be installed).

Protobuf

To compile protobuf, in biomaj_download/message:

protoc --python_out=. downmessage.proto

Development

flake8 --ignore E501 biomaj_download/\*.py biomaj_download/download

Test

To run the test suite, use:

nosetests -a '!local_irods' tests/biomaj_tests.py

This command skips the test that need a local iRODS server.

Some test might fail due to network connection. You can skip them with:

nosetests -a '!network' tests/biomaj_tests.py

(To skip the local iRODS test and the network tests, use -a '!network,!local_irods').

Run

Message consumer:

export BIOMAJ_CONFIG=path_to_config.yml
python bin/biomaj_download_consumer.py

Web server

If package is installed via pip, you need a file named gunicorn_conf.py containing somehwhere on local server:

def worker_exit(server, worker):
    from prometheus_client import multiprocess
    multiprocess.mark_process_dead(worker.pid)

If you cloned the repository and installed it via python setup.py install, just refer to the gunicorn_conf.py in the cloned repository.

export BIOMAJ_CONFIG=path_to_config.yml
rm -rf ..path_to/prometheus-multiproc
mkdir -p ..path_to/prometheus-multiproc
export prometheus_multiproc_dir=..path_to/prometheus-multiproc
gunicorn -c gunicorn_conf.py biomaj_download.biomaj_download_web:app

Web processes should be behind a proxy/load balancer, API base url /api/download

Prometheus endpoint metrics are exposed via /metrics on web server

Retrying

A common problem when downloading a large number of files is the handling of temporary failures (network issues, server too busy to answer, etc.). Since version 3.1.2, biomaj-download uses the Tenacity library which is designed to handle this. This mechanism is configurable through 2 downloader-specific options (see Download options): stop_condition and wait_policy.

When working on python code, you can pass instances of Tenacity's stop_base and wait_base respectively. This includes classes defined in Tenacity or your own derived classes.

For bank configuration those options also parse strings read from the configuration file. This parsing is based on the Simple Eval library. The rules are straightforward:

• All concrete stop and wait classes defined in Tenacity (i.e. classes inheriting from stop_base and wait_base respectively) can be used by calling their constructor with the expected parameters. For example, the string "stop_after_attempt(5)" will create the desired object. Note that stop and wait classes that need no argument must be used as constants (i.e. use "stop_never" and not "stop_never()"). Currently, this is the case for "stop_never" (as in Tenacity) and "wait_none" (this slightly differs from Tenacity where it is "wait_none()").
• You can use classes that allow to combine other stop conditions (namely stop_all and stop_any) or wait policies (namely wait_combine).
• Operator + can be used to add wait policies (similar to wait_combine).
• Operators & and | can be used to compose stop conditions (similar to wait_all and wait_none respectively).

However, in this case, you can't use your own conditions. The complete list of stop conditions is:

• stop_never (although its use is discouraged)
• stop_after_attempt
• stop_after_delay
• stop_when_event_set
• stop_all
• stop_any

The complete list of wait policies is:

• wait_none
• wait_fixed
• wait_random
• wait_incrementing
• wait_exponential
• wait_random_exponential
• wait_combine
• wait_chain

Please refer to Tenacity doc for their meaning and their parameters.

Examples (inspired by Tenacity doc):

• "wait_fixed(3) + wait_random(0, 2)" and "wait_combine(wait_fixed(3), wait_random(0, 2))" are equivalent and will wait 3 seconds + up to 2 seconds of random delay
• "wait_chain(*([wait_fixed(3) for i in range(3)] + [wait_fixed(7) for i in range(2)] + [wait_fixed(9)]))" will wait 3s for 3 attempts, 7s for the next 2 attempts and 9s for all attempts thereafter (here + is the list concatenation).
• "wait_none + wait_random(1,2)" will wait between 1s and 2s (since wait_none doesn't wait).
• "stop_never | stop_after_attempt(5)" will stop after 5 attempts (since stop_never never stops).

Note that some protocols (e.g. FTP) classify errors as temporary or permanent (for example trying to download inexisting file). More generally, we could distinguish permanent errors based on error codes, etc. and not retry in this case. However in our experience, so called permanent errors may well be temporary. Therefore downloaders always retry whatever the error. In some cases, this is a waste of time but generally this is worth it.

Host keys

When using the sftp protocol, biomaj-download must check the host key. Those keys are stored in a file (for instance ~/.ssh/known_hosts).

Two options are available to configure this:

• ssh_hosts_file which sets the file to use
• ssh_new_host which sets what to do for a new host

When the host and the key are found in the file, the connection is accepted. If the host is found but the key missmatches, the connection is rejected (this usually indicates a problem or a change of configuration on the remote server). When the host is not found, the decision depends on the value of ssh_new_host:

• reject means that the connection is rejected
• accept means that the connection is accepted
• add means that the connection is accepted and the key is added to the file

See the description of the options in Download options.

Download options

Since version 3.0.26, you can use the set_options method to pass a dictionary of downloader-specific options. The following list shows some options and their effect (the option to set is the key and the parameter is the associated value):

• stop_condition:
  • parameter: an instance of Tenacity stop_base or a string (see Retrying).
  • downloader(s): all (except LocalDownload).
  • effect: sets the condition on which we should stop retrying to download a file.
  • default: stop_after_attempt(3) (i.e. stop after 3 attempts).
• wait_policy:
  • parameter: an instance of Tenacity wait_base or a string (see Retrying).
  • downloader(s): all (except LocalDownload).
  • effect: sets the wait policy between download attempts.
  • default: wait_fixed(3) (i.e. wait 3 seconds between attempts).
• skip_check_uncompress:
  • parameter: bool.
  • downloader(s): all (except LocalDownload).
  • effect: if true, don't test the archives after download.
  • default: false (i.e. test the archives).
• ssl_verifyhost:
  • parameter: bool.
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload).
  • effect: if false, don't check that the name of the remote server is the same than in the SSL certificate.
  • default: true (i.e. check host name).
  • note: it's generally a bad idea to disable this verification. However some servers are badly configured. See here for the corresponding cURL option.
• ssl_verifypeer:
  • parameter: bool.
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload).
  • effect: if false, don't check the authenticity of the peer's certificate.
  • default: true (i.e. check authenticity).
  • note: it's generally a bad idea to disable this verification. However some servers are badly configured. See here for the corresponding cURL option.
• ssl_server_cert:
  • parameter: path of the certificate file.
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload).
  • effect: use the certificate(s) in this file to verify the peer with.
  • default: use OS certificates.
  • note: see here for the corresponding cURL option.
  • parameter: int.
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload).
  • effect: sets the interval, in seconds, that the operating system will wait between sending keepalive probes.
  • default: cURL default (60s at the time of this writing).
  • note: see here for the corresponding cURL option.
• ftp_method:
  • parameter: one of default, multicwd, nocwd, singlecwd (case insensitive).
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload) - only used for FTP(S).
  • effect: sets the method used to reach a file on a FTP(S) server (nocwd and singlecwd are usually faster but not always supported).
  • default: default (which is multicwd at the time of this writing as in cURL).
  • note: see here for the corresponding cURL option.
• ssh_hosts_file:
  • parameter: path of the known hosts file.
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload) - only used for SFTP.
  • effect: sets the file used to read/store host keys for SFTP.
  • default: ~/.ssh/known_hosts (where ~ is the home directory of the current user).
  • note: see here for the corresponding cURL option and the option below.
• ssh_new_host:
  • parameter: one of reject, accept, add.
  • downloader(s): CurlDownload (and derived classes: DirectFTPDownload, DirectHTTPDownload) - only used for SFTP.
  • effect: sets the policy to use for an unknown host.
  • default: reject (i.e. refuse new hosts - you must add them in the file for instance with ssh or sftp).
  • note: see here for the corresponding cURL option and the option above.

Those options can be set in bank properties. See file global.properties.example in biomaj module.

3.2.1: #26 Accept new keys for SFTP servers (Closes #25) Strip extra slash characters in remote file list (due to regexp parsing) #20 Add a configurable mechanism to retry download when it fails #24 Speed up IRODDownload Introduce a method to perform configuration before network methods. Adapt implementation of generic methods and subclasses. Resolve bug when the parser analyse also the Message Of The Day when it wants only list of file. (#23) 3.1.2: #18 Add a protocol option to set CURLOPT_FTP_FILEMETHOD #19 Rename protocol options to options Fix copy of production files instead of download when files are in subdirectories 3.1.1: #17 Support MDTM command in directftp 3.1.0: #16 Don't change name after download in DirectHTTPDownloader PR #7 Refactor downloaders (WARNING breaks API) 3.0.27: Fix previous release broken with a bug in direct protocols 3.0.26: Change default download timeout to 1h #12 Allow FTPS protocol #14 Add mechanism for protocol specific options 3.0.25: Allow to use hardlinks in LocalDownload 3.0.24: Remove debug logs 3.0.23: Support spaces in remote file names 3.0.22: Fix */ remote.files parsing 3.0.21: Fix traefik labels 3.0.20: Update pika dependency release Add tags for traefik support 3.0.19: Check archives after download Fix python regexps syntax (deprecation) 3.0.18: Rename protobuf and use specific package to avoid conflicts 3.0.17: Regenerate protobuf message desc, failing on python3 3.0.16: Add missing req in setup.py 3.0.15: Fix progress download control where could have infinite loop Add irods download

3.0.14: Allow setup of local_endpoint per service, else use default local_endpoint

3.0.13: In rate limiting, add progress vs total of download Fix rate limiting submission

3.0.12: Add retry in case of session creation failure disable web thread logging

3.0.11: Display progress of download by percent of downloads In case of contact error in downloadclient, retry connection

3.0.10: Feature #3: Add rate limiting option to limit number of parallel downloads for a client

3.0.9: Add host in prometheus stats Fix #2: allow setting http.group.file.size or http.group.file.date to -1 if not avalaible in http(s) page for regexp

3.0.8: Fix prometheus stats Add consul supervision

3.0.7: Change size type to int64

3.0.6: Fix download_or_copy to avoid downloading a file existing in a previous production directory

3.0.4: Fixes on messages

3.0.3: Fix management of timeout leading to a crash when using biomaj.download parameter.

3.0.2: set rabbitmq parameter optional

3.0.1: add missing README etc.. in package

3.0.0: move download management out of biomaj main package