path: root/bh20simplewebuploader/main.py

import collections
import tempfile
import shutil
import subprocess
import os
import sys
import re
import string
import yaml
import pkg_resources
from flask import Flask, request, redirect, send_file, send_from_directory, render_template

import os.path

if not os.path.isfile('bh20sequploader/mainx.py'):
    print("WARNING: run FLASK from the root of the source repository!", file=sys.stderr)

app = Flask(__name__, static_url_path='/static', static_folder='static')

# Limit file upload size. We shouldn't be working with anything over 1 MB; these are small genomes.
# We will enforce the limit ourselves and set a higher safety limit here.
app.config['MAX_CONTENT_LENGTH'] = 50 * 1024 * 1024

# When a file is too big we get a 413.
@app.errorhandler(413)
def handle_large_file(e):
    return (render_template('error.html',
        error_message="One of your files is too large. The maximum file size is 50 megabytes."), 413)


def type_to_heading(type_name):
    """
    Turn a type name like "sampleSchema" from the metadata schema into a human-readable heading.
    """

    # Remove camel case
    decamel = re.sub('([A-Z])', r' \1', type_name)
    # Split
    parts = decamel.split()
    # Capitalize words and remove unwanted components
    filtered = [part.capitalize() for part in parts if (part.lower() != 'schema' and part != '')]
    # Reassemble
    return ' '.join(filtered)

def name_to_label(field_name):
    """
    Turn a filed name like "host_health_status" from the metadata schema into a human-readable label.
    """

    return string.capwords(field_name.replace('_', ' '))

def is_iri(string):
    """
    Return True if the given string looks like an IRI, and False otherwise.

    Used for finding type IRIs in the schema.

    Right now only supports http(s) URLs because that's all we have in our schema.
    """

    return string.startswith('http')

def generate_form(schema, options):
    """
    Linearize the schema into a list of dicts.

    Each dict either has a 'heading' (in which case we put a heading for a
    form section in the template) or an 'id', 'label', 'type', and 'required'
    (in which case we make a form field in the template). Non-heading dicts
    with type 'select' will have an 'options' field, with a list of (name,
    value) tuples, and represent a form dropdown element. Non-heading dicts may
    have a human-readable 'docstring' field describing them.

    Takes the deserialized metadata schema YAML, and also a deserialized YAML
    of option values. The option values are keyed on (unscoped) field name in
    the schema, and each is a dict of human readable option -> corresponding
    IRI.
    """

    # Get the list of form components, one of which is the root
    components = schema.get('$graph', [])

    # Find the root
    root_name = None
    # And also index components by type name
    by_name = {}
    for component in components:
        # Get the name of each
        component_name = component.get('name', None)
        if isinstance(component_name, str):
            # And remember how to map back form it
            by_name[component_name] = component
        if component.get('documentRoot', False):
            # Find whichever one is the root
            root_name = component_name


    def walk_fields(type_name, parent_keys=['metadata'], subtree_optional=False):
        """
        Do a traversal of the component tree.
        Yield a bunch of form item dicts, in order.
        Form IDs are .-separated keypaths for where they are in the structure.
        parent_keys is the path of field names to where we are in the root record's document tree.
        """

        if len(parent_keys) > 1:
            # First make a heading, if we aren't the very root of the form
            yield {'heading': type_to_heading(type_name)}

        for field_name, field_type in by_name.get(type_name, {}).get('fields', {}).items():
            # For each field

            ref_iri = None
            docstring = None
            if not isinstance(field_type, str):
                # If the type isn't a string
                
                # It may have documentation
                docstring = field_type.get('doc', None)

                # See if it has a more info/what goes here URL
                predicate = field_type.get('jsonldPredicate', {})
                # Predicate may be a URL, a dict with a URL in _id, maybe a
                # dict with a URL in _type, or a dict with _id and _type but no
                # URLs anywhere. Some of these may not technically be allowed
                # by the format, but if they occur, we might as well try to
                # handle them.
                if isinstance(predicate, str):
                    if is_iri(predicate):
                        ref_iri = predicate
                else:
                    # Assume it's a dict. Look at the fields we know about.
                    for field in ['_id', 'type']:
                        field_value = predicate.get(field, None)
                        if isinstance(field_value, str) and is_iri(field_value) and ref_iri is None:
                            # Take the first URL-looking thing we find
                            ref_iri = field_value
                            break


                # Now overwrite the field type with the actual type string
                field_type = field_type.get('type', '')

            # Decide if the field is optional (type ends in ?)
            optional = False
            if len(field_type) > 0 and field_type[-1] == '?':
                # It's optional
                optional = True
                # Drop the ?
                field_type = field_type[:-1]

            if field_type in by_name:
                # This is a subrecord. We need to recurse
                for item in walk_fields(field_type, parent_keys + [field_name], subtree_optional or optional):
                    yield item
            else:
                # This is a leaf field. We need an input for it.
                record = {}
                record['id'] = '.'.join(parent_keys + [field_name])
                record['label'] = name_to_label(field_name)
                record['required'] = not optional and not subtree_optional
                if ref_iri:
                    record['ref_iri'] = ref_iri
                if docstring:
                    record['docstring'] = docstring

                if field_name in options:
                    # The field will be a 'select' type no matter what its real
                    # data type is.
                    record['type'] = 'select' # Not a real HTML input type. It's its own tag.
                    # We have a set of values to present
                    record['options'] = []
                    for name, value in options[field_name].items():
                        # Make a tuple for each one
                        record['options'].append((name, value))
                elif field_type == 'string':
                    record['type'] = 'text' # HTML input type
                elif field_type == 'int':
                    record['type'] = 'number'
                else:
                    raise NotImplementedError('Unimplemented field type {} in {} in metadata schema'.format(field_type, type_name))
                yield record

    return list(walk_fields(root_name))


# At startup, we need to load the metadata schema from the uploader module, so we can make a form for it
METADATA_SCHEMA = yaml.safe_load(pkg_resources.resource_stream("bh20sequploader", "bh20seq-schema.yml"))
METADATA_OPTION_DEFINITIONS = yaml.safe_load(pkg_resources.resource_stream("bh20sequploader", "bh20seq-options.yml"))
FORM_ITEMS = generate_form(METADATA_SCHEMA, METADATA_OPTION_DEFINITIONS)

@app.route('/')
def send_form():
    """
    Send the file upload form/front page.
    """

    return render_template('form.html', fields=FORM_ITEMS)

class FileTooBigError(RuntimeError):
    """
    Raised when the user gives a file that is too large.
    """
    pass

def copy_with_limit(in_file, out_file, limit=1024*1024):
    """
    Copy a file stream, and raise FileTooBigError if the file is too big.
    """

    bytes_used = 0
    buf_size = 65536

    buf = in_file.read(buf_size)
    bytes_used += len(buf)
    while buf:
        if bytes_used > limit:
            raise FileTooBigError('Hit file length limit')
        out_file.write(buf)
        buf = in_file.read(buf_size)
        bytes_used += len(buf)

def parse_input(input_string, html_type):
    """
    Parse an input from the given HTML input type into a useful Python type.

    Raise ValueError if something does not parse.
    Raise NotImplementedError if we forgot to implement a type.
    """

    if html_type == 'text':
        return input_string
    elif html_type == 'number':
        return int(input_string)
    else:
        raise NotImplementedError('Unimplemented input type: {}'.format(html_type))

@app.route('/submit', methods=['POST'])
def receive_files():
    """
    Receive the uploaded files.
    """

    # We're going to work in one directory per request
    dest_dir = tempfile.mkdtemp()
    # The uploader will happily accept a FASTQ with this name
    fasta_dest = os.path.join(dest_dir, 'fasta.fa')
    metadata_dest = os.path.join(dest_dir, 'metadata.json')
    try:
        if 'fasta' not in request.files:
            return (render_template('error.html',
                error_message="You did not include a FASTA or FASTQ file."), 403)
        try:
            with open(fasta_dest, 'wb') as out_stream:
                # Use a plausible file size limit for a little FASTQ
                copy_with_limit(request.files.get('fasta').stream, out_stream, limit=50*1024*1024)
        except FileTooBigError as e:
            # Delegate to the 413 error handler
            return handle_large_file(e)

        if request.form.get('metadata_type', None) == 'upload':
            if 'metadata' not in request.files:
                return (render_template('error.html',
                    error_message="You did not include a metadata file."), 403)
            try:
                with open(metadata_dest, 'wb') as out_stream:
                    copy_with_limit(request.files.get('metadata').stream, out_stream)
            except FileTooBigError as e:
                # Delegate to the 413 error handler
                return handle_large_file(e)
        elif request.form.get('metadata_type', None) == 'fill':
            # Build a metadata dict
            metadata = {}

            for item in FORM_ITEMS:
                # Pull all the field values we wanted from the form
                if 'heading' in item:
                    continue

                if item['id'] in request.form and len(request.form[item['id']]) > 0:
                    # We have this thing. Make a place in the dict tree for it.
                    parts = item['id'].split('.')
                    key = parts[-1]
                    # Remove leading 'metadata'
                    path = parts[1:-1]
                    dest_dict = metadata
                    for parent in path:
                        if parent not in dest_dict:
                            dest_dict[parent] = {}
                        dest_dict = dest_dict[parent]

                    try:
                        # Now finally add the item
                        dest_dict[key] = parse_input(request.form[item['id']], item['type'])
                    except ValueError:
                        # We don't like that input
                        return (render_template('error.html',
                            error_message="You provided an unacceptable value for the metadata item {}".format(item['id'])), 403)
                elif item['required']:
                    return (render_template('error.html',
                            error_message="You omitted the required metadata item {}".format(item['id'])), 403)

            # Now serialize the file with all the items
            with open(metadata_dest, 'w') as out_stream:
                yaml.dump(metadata, out_stream)
        else:
            return (render_template('error.html',
                    error_message="You did not include metadata."), 403)

        # Try and upload files to Arvados using the sequence uploader CLI

        cmd = ['python3','bh20sequploader/main.py', fasta_dest, metadata_dest]
        print(" ".join(cmd),file=sys.stderr)
        result = subprocess.run(cmd,
            stdout=subprocess.PIPE, stderr=subprocess.PIPE)

        if result.returncode != 0:
            # It didn't work. Complain.
            error_message="Uploader returned value {} and said:".format(result.returncode) + str(result.stderr.decode('utf-8'))
            print(error_message, file=sys.stderr)
            return (render_template('error.html', error_message=error_message), 403)
        else:
            # It worked. Say so.
            return render_template('success.html', log=result.stdout.decode('utf-8', errors='replace'))
    finally:
        shutil.rmtree(dest_dir)