zulip/zerver/management/commands/export_search.py

264 lines
10 KiB
Python

import csv
import os
import shutil
from argparse import ArgumentParser
from datetime import datetime, timezone
from email.headerregistry import Address
from functools import lru_cache, reduce
from operator import or_
from typing import Any
import orjson
from django.core.management.base import CommandError
from django.db.models import Q
from typing_extensions import override
from zerver.lib.management import ZulipBaseCommand
from zerver.lib.soft_deactivation import reactivate_user_if_soft_deactivated
from zerver.lib.upload import save_attachment_contents
from zerver.models import AbstractUserMessage, Attachment, Message, Recipient, Stream, UserProfile
from zerver.models.recipients import get_or_create_direct_message_group
from zerver.models.users import get_user_by_delivery_email
def write_attachment(base_path: str, attachment: Attachment) -> None:
dir_path_id = os.path.dirname(attachment.path_id)
assert "../" not in dir_path_id
os.makedirs(base_path + "/" + dir_path_id, exist_ok=True)
with open(base_path + "/" + attachment.path_id, "wb") as attachment_file:
save_attachment_contents(attachment.path_id, attachment_file)
class Command(ZulipBaseCommand):
help = """Exports the messages matching certain search terms, or from
senders/recipients.
This is most often used for legal compliance.
"""
@override
def add_arguments(self, parser: ArgumentParser) -> None:
self.add_realm_args(parser, required=True)
parser.add_argument(
"--output",
metavar="<path>",
help="File to output JSON/CSV results to; it must not exist, unless --force is given",
required=True,
)
parser.add_argument(
"--write-attachments",
metavar="<directory>",
help="If provided, export all referenced attachments into the directory",
)
parser.add_argument(
"--force", action="store_true", help="Overwrite the output file if it exists already"
)
parser.add_argument(
"--file",
metavar="<path>",
help="Read search terms from the named file, one per line",
)
parser.add_argument(
"search_terms",
nargs="*",
metavar="<search term>",
help="Terms to search for in message body or topic",
)
parser.add_argument(
"--after",
metavar="<datetime>",
help="Limit to messages on or after this ISO datetime, treated as UTC",
type=lambda s: datetime.fromisoformat(s).astimezone(timezone.utc),
)
parser.add_argument(
"--before",
metavar="<datetime>",
help="Limit to messages on or before this ISO datetime, treated as UTC",
type=lambda s: datetime.fromisoformat(s).astimezone(timezone.utc),
)
users = parser.add_mutually_exclusive_group()
users.add_argument(
"--sender",
action="append",
metavar="<email>",
help="Limit to messages sent by users with any of these emails (may be specified more than once)",
)
users.add_argument(
"--recipient",
action="append",
metavar="<email>",
help="Limit to messages received by users with any of these emails (may be specified more than once). This is a superset of --sender, since senders receive every message they send.",
)
users.add_argument(
"--dm",
action="append",
metavar="<email>",
help="Limit to messages in a DM between all of the users provided.",
)
@override
def handle(self, *args: Any, **options: Any) -> None:
terms = set()
if options["file"]:
with open(options["file"]) as f:
terms.update(f.read().splitlines())
terms.update(options["search_terms"])
if (
not terms
and not options["before"]
and not options["after"]
and not options["sender"]
and not options["recipient"]
and not options["dm"]
):
raise CommandError("One or more limits are required!")
if not options["output"].endswith((".json", ".csv")):
raise CommandError(
"Unknown file format: {options['output']} Only .csv and .json are supported"
)
if os.path.exists(options["output"]) and not options["force"]:
raise CommandError(
f"Output path '{options['output']}' already exists; use --force to overwrite"
)
if options["write_attachments"] and os.path.exists(options["write_attachments"]):
if not options["force"]:
raise CommandError(
f"Attachments output path '{options['write_attachments']}' already exists; use --force to overwrite"
)
shutil.rmtree(options["write_attachments"])
realm = self.get_realm(options)
assert realm is not None
limits = Q()
limits = reduce(
or_,
[Q(content__icontains=term) | Q(subject__icontains=term) for term in terms],
limits,
)
if options["after"]:
limits &= Q(date_sent__gt=options["after"])
if options["before"]:
limits &= Q(date_sent__lt=options["before"])
if options["recipient"]:
user_profiles = [get_user_by_delivery_email(e, realm) for e in options["recipient"]]
for user_profile in user_profiles:
# Users need to not be long-term idle for the
# UserMessages to be a judge of which messages they
# received.
reactivate_user_if_soft_deactivated(user_profile)
limits &= Q(
usermessage__user_profile_id__in=[user_profile.id for user_profile in user_profiles]
)
elif options["sender"]:
limits &= reduce(
or_,
[Q(sender__delivery_email__iexact=e) for e in options["sender"]],
)
elif options["dm"]:
user_profiles = [get_user_by_delivery_email(e, realm) for e in options["dm"]]
for user_profile in user_profiles:
reactivate_user_if_soft_deactivated(user_profile)
if len(user_profiles) == 1:
limits &= Q(
usermessage__user_profile_id=user_profiles[0].id,
usermessage__flags__andnz=AbstractUserMessage.flags.is_private.mask,
)
elif len(user_profiles) == 2:
user_a, user_b = user_profiles
limits &= Q(recipient=user_a.recipient, sender=user_b) | Q(
recipient=user_b.recipient, sender=user_a
)
else:
direct_message_group = get_or_create_direct_message_group(
[user.id for user in user_profiles]
)
limits &= Q(recipient=direct_message_group.recipient)
attachments_written: set[str] = set()
messages_query = Message.objects.filter(limits, realm=realm).order_by("date_sent")
print(f"Exporting {len(messages_query)} messages...")
@lru_cache(maxsize=1000)
def format_sender(full_name: str, delivery_email: str) -> str:
return str(Address(display_name=full_name, addr_spec=delivery_email))
def format_full_recipient(recipient_id: int, subject: str) -> str:
recip_str, has_subject = format_recipient(recipient_id)
if not has_subject:
return recip_str
return f"{recip_str} > {subject}"
@lru_cache(maxsize=1000)
def format_recipient(recipient_id: int) -> tuple[str, bool]:
recipient = Recipient.objects.get(id=recipient_id)
if recipient.type == Recipient.STREAM:
stream = Stream.objects.values("name").get(id=recipient.type_id)
return "#" + stream["name"], True
users = (
UserProfile.objects.filter(
subscription__recipient_id=recipient.id,
)
.order_by("full_name")
.values_list("full_name", "delivery_email")
)
return ", ".join(format_sender(e[0], e[1]) for e in users), False
def transform_message(message: Message) -> dict[str, str]:
row = {
"id": str(message.id),
"timestamp (UTC)": message.date_sent.astimezone(timezone.utc).strftime(
"%Y-%m-%d %H:%M:%S"
),
"sender": format_sender(message.sender.full_name, message.sender.delivery_email),
"recipient": format_full_recipient(message.recipient_id, message.subject),
"content": message.content,
"edit history": message.edit_history if message.edit_history is not None else "",
}
if options["write_attachments"]:
if message.has_attachment:
attachments = message.attachment_set.all()
row["attachments"] = " ".join(a.path_id for a in attachments)
for attachment in attachments:
if attachment.path_id in attachments_written:
continue
write_attachment(options["write_attachments"], attachment)
attachments_written.add(attachment.path_id)
else:
row["attachments"] = ""
return row
if options["output"].endswith(".json"):
with open(options["output"], "wb") as json_file:
json_file.write(
# orjson doesn't support dumping from a generator
orjson.dumps(
[transform_message(m) for m in messages_query], option=orjson.OPT_INDENT_2
)
)
elif options["output"].endswith(".csv"):
with open(options["output"], "w") as csv_file:
columns = [
"id",
"timestamp (UTC)",
"sender",
"recipient",
"content",
"edit history",
]
if options["write_attachments"]:
columns += ["attachments"]
csvwriter = csv.DictWriter(csv_file, columns)
csvwriter.writeheader()
csvwriter.writerows(transform_message(m) for m in messages_query)