271 lines
9.6 KiB
Python
Raw Normal View History

2023-09-02 08:50:31 -07:00
"""
Salesforce Connector
Able to download Account, Case, Campaign, EmailMessage, Lead
Salesforce returns everything as a list of json.
This saves each entry as a separate file to be partitioned.
Using JWT authorization
https://developer.salesforce.com/docs/atlas.en-us.sfdx_dev.meta/sfdx_dev/sfdx_dev_auth_key_and_cert.htm
https://developer.salesforce.com/docs/atlas.en-us.sfdx_dev.meta/sfdx_dev/sfdx_dev_auth_connected_app.htm
"""
import os
import typing as t
from collections import OrderedDict
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
from dataclasses import dataclass, field
from datetime import datetime
2023-09-02 08:50:31 -07:00
from email.utils import formatdate
from pathlib import Path
from string import Template
from textwrap import dedent
from dateutil import parser # type: ignore
from unstructured.ingest.error import SourceConnectionError, SourceConnectionNetworkError
2023-09-02 08:50:31 -07:00
from unstructured.ingest.interfaces import (
BaseConnectorConfig,
BaseIngestDoc,
BaseSourceConnector,
2023-09-02 08:50:31 -07:00
IngestDocCleanupMixin,
SourceConnectorCleanupMixin,
SourceMetadata,
2023-09-02 08:50:31 -07:00
)
from unstructured.ingest.logger import logger
from unstructured.utils import requires_dependencies
class MissingCategoryError(Exception):
"""There are no categories with that name."""
SALESFORCE_API_VERSION = "57.0"
2023-09-02 08:50:31 -07:00
ACCEPTED_CATEGORIES = ["Account", "Case", "Campaign", "EmailMessage", "Lead"]
2023-09-02 08:50:31 -07:00
EMAIL_TEMPLATE = Template(
"""MIME-Version: 1.0
Date: $date
Message-ID: $message_identifier
Subject: $subject
From: $from_email
To: $to_email
Content-Type: multipart/alternative; boundary="00000000000095c9b205eff92630"
--00000000000095c9b205eff92630
Content-Type: text/plain; charset="UTF-8"
$textbody
--00000000000095c9b205eff92630
Content-Type: text/html; charset="UTF-8"
$htmlbody
2023-09-02 08:50:31 -07:00
--00000000000095c9b205eff92630--
""",
)
@dataclass
class SimpleSalesforceConfig(BaseConnectorConfig):
"""Connector specific attributes"""
categories: t.List[str]
2023-09-02 08:50:31 -07:00
username: str
consumer_key: str
private_key_path: str
recursive: bool = False
@requires_dependencies(["simple_salesforce"], extras="salesforce")
def get_client(self):
2023-09-02 08:50:31 -07:00
from simple_salesforce import Salesforce
return Salesforce(
username=self.username,
consumer_key=self.consumer_key,
privatekey_file=self.private_key_path,
version=SALESFORCE_API_VERSION,
2023-09-02 08:50:31 -07:00
)
@dataclass
class SalesforceIngestDoc(IngestDocCleanupMixin, BaseIngestDoc):
connector_config: SimpleSalesforceConfig
2023-09-02 08:50:31 -07:00
record_type: str
record_id: str
registry_name: str = "salesforce"
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
_record: OrderedDict = field(default_factory=lambda: OrderedDict())
@property
def record(self):
if not self._record:
self._record = self.get_record()
return self._record
2023-09-02 08:50:31 -07:00
def _tmp_download_file(self) -> Path:
if self.record_type == "EmailMessage":
record_file = self.record_id + ".eml"
elif self.record_type in ["Account", "Lead", "Case", "Campaign"]:
record_file = self.record_id + ".xml"
2023-09-02 08:50:31 -07:00
else:
raise MissingCategoryError(
f"There are no categories with the name: {self.record_type}",
)
return Path(self.read_config.download_dir) / self.record_type / record_file
2023-09-02 08:50:31 -07:00
@property
def _output_filename(self) -> Path:
record_file = self.record_id + ".json"
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
return Path(self.processor_config.output_dir) / self.record_type / record_file
2023-09-02 08:50:31 -07:00
def _create_full_tmp_dir_path(self):
self._tmp_download_file().parent.mkdir(parents=True, exist_ok=True)
def _xml_for_record(self, record: OrderedDict) -> str:
"""Creates partitionable xml file from a record"""
import xml.etree.ElementTree as ET
def flatten_dict(data, parent, prefix=""):
for key, value in data.items():
if isinstance(value, OrderedDict):
flatten_dict(value, parent, prefix=f"{prefix}{key}.")
else:
item = ET.Element("item")
item.text = f"{prefix}{key}: {value}"
parent.append(item)
root = ET.Element("root")
flatten_dict(record, root)
xml_string = ET.tostring(root, encoding="utf-8", xml_declaration=True).decode()
return xml_string
def _eml_for_record(self, email_json: t.Dict[str, t.Any]) -> str:
2023-09-02 08:50:31 -07:00
"""Recreates standard expected .eml format using template."""
eml = EMAIL_TEMPLATE.substitute(
date=formatdate(parser.parse(email_json.get("MessageDate")).timestamp()),
message_identifier=email_json.get("MessageIdentifier"),
subject=email_json.get("Subject"),
from_email=email_json.get("FromAddress"),
to_email=email_json.get("ToAddress"),
textbody=email_json.get("TextBody"),
# TODO: This is a hack to get emails to process correctly.
# The HTML partitioner seems to have issues with <br> and text without tags like <p>
htmlbody=email_json.get("HtmlBody", "") # "" because you can't .replace None
.replace("<br />", "<p>")
.replace("<body", "<body><p"),
2023-09-02 08:50:31 -07:00
)
return dedent(eml)
@SourceConnectionNetworkError.wrap
def _get_response(self):
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
client = self.connector_config.get_client()
return client.query_all(
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
f"select FIELDS(STANDARD) from {self.record_type} where Id='{self.record_id}'",
)
def get_record(self) -> OrderedDict:
# Get record from Salesforce based on id
response = self._get_response()
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
logger.debug(f"response from salesforce record request: {response}")
records = response["records"]
if not records:
raise ValueError(f"No record found with record id {self.record_id}: {response}")
record_json = records[0]
return record_json
def update_source_metadata(self) -> None: # type: ignore
record_json = self.record
date_format = "%Y-%m-%dT%H:%M:%S.000+0000"
self.source_metadata = SourceMetadata(
date_created=datetime.strptime(record_json["CreatedDate"], date_format).isoformat(),
date_modified=datetime.strptime(
record_json["LastModifiedDate"],
date_format,
).isoformat(),
# SystemModstamp is Timestamp if record has been modified by person or automated system
version=record_json.get("SystemModstamp"),
source_url=record_json["attributes"].get("url"),
exists=True,
)
@SourceConnectionError.wrap
2023-09-02 08:50:31 -07:00
@BaseIngestDoc.skip_if_file_exists
def get_file(self):
"""Saves individual json records locally."""
self._create_full_tmp_dir_path()
logger.debug(f"Writing file {self.record_id} - PID: {os.getpid()}")
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
record = self.record
2023-09-02 08:50:31 -07:00
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
self.update_source_metadata()
2023-09-02 08:50:31 -07:00
try:
if self.record_type == "EmailMessage":
document = self._eml_for_record(record)
else:
document = self._xml_for_record(record)
2023-09-02 08:50:31 -07:00
with open(self._tmp_download_file(), "w") as page_file:
page_file.write(document)
2023-09-02 08:50:31 -07:00
except Exception as e:
logger.error(
f"Error while downloading and saving file: {self.record_id}.",
)
logger.error(e)
@property
def filename(self):
"""The filename of the file created from a Salesforce record"""
return self._tmp_download_file()
@dataclass
class SalesforceSourceConnector(SourceConnectorCleanupMixin, BaseSourceConnector):
connector_config: SimpleSalesforceConfig
2023-10-17 12:15:08 -04:00
def __post_init__(self):
self.ingest_doc_cls: t.Type[SalesforceIngestDoc] = SalesforceIngestDoc
2023-09-02 08:50:31 -07:00
def initialize(self):
pass
@requires_dependencies(["simple_salesforce"], extras="salesforce")
def check_connection(self):
from simple_salesforce.exceptions import SalesforceError
try:
self.connector_config.get_client()
except SalesforceError as salesforce_error:
logger.error(f"failed to validate connection: {salesforce_error}", exc_info=True)
raise SourceConnectionError(f"failed to validate connection: {salesforce_error}")
2023-09-02 08:50:31 -07:00
@requires_dependencies(["simple_salesforce"], extras="salesforce")
def get_ingest_docs(self) -> t.List[SalesforceIngestDoc]:
2023-09-02 08:50:31 -07:00
"""Get Salesforce Ids for the records.
Send them to next phase where each doc gets downloaded into the
appropriate format for partitioning.
"""
from simple_salesforce.exceptions import SalesforceMalformedRequest
client = self.connector_config.get_client()
2023-09-02 08:50:31 -07:00
ingest_docs = []
for record_type in self.connector_config.categories:
2023-09-02 08:50:31 -07:00
if record_type not in ACCEPTED_CATEGORIES:
raise ValueError(f"{record_type} not currently an accepted Salesforce category")
try:
# Get ids from Salesforce
records = client.query_all(
f"select Id from {record_type}",
)
for record in records["records"]:
ingest_docs.append(
SalesforceIngestDoc(
connector_config=self.connector_config,
refactor: unstructured ingest as a pipeline (#1551) ### Description As we add more and more steps to the pipeline (i.e. chunking, embedding, table manipulation), it would help seperate the responsibility of each of these into their own processes, running each in parallel using json files to share data across. This will also help guarantee data is serializable if this code was used in an actual pipeline. Following is a flow diagram of the proposed changes. As part of this change: * A parent pipeline class will be responsible for running each `node`, which can optionally be run via multiprocessing if it supports it, or not. Possible nodes at this moment: * Doc factory: creates all the ingest docs via the source connector * Source: reads/downloads all of the content to process to the local filesystem to the location set by the `download_dir` parameter. * Partition: runs partition on all of the downloaded content in json format. * Any number of reformat nodes that modify the partitioned content. This can include chunking, embedding, etc. * Write: push the final json into the destination via the destination connector * This pipeline relies on the information of the ingest docs to be available via their serialization. An optimization was introduced with the `IngestDocJsonMixin` which adds in all the `@property` fields to the serialized json already being created via the `DataClassJsonMixin` * For all intermediate steps (partitioning, reformatting), the content is saved to a dedicated location on the local filesystem. Right now it's set to `$HOME/.cache/unstructured/ingest/pipeline/STEP_NAME/`. * Minor changes: made sense to move some of the config parameters between the read and partition configs when I explicitly divided the responsibility to download vs partition the content in the pipeline. * The pipeline class only makes the doc factory, source and partition nodes required, keeping with the logic that has been supported so far. All reformatting nodes and write node are optional. * Long term, there should also be some changes to the base configs supported by the CLI to support pipeline specific configs, but for now what exists was used to minimize changes in this PR. * Final step to copy the final output to the location designated by the `_output_filename` value of the ingest doc. * Hashing occurs at each step by hashing the parameters of that step (i.e. partition configs) along with the previous step via the filename used. This allows each step to be the same _if_ all the parameters for it have not changed and the content so far is the same. * The only data that is shared and has writes to across processes is the dictionary of ingest json data. This dict is created using the `multiprocessing.manager.DictProxy` to make sure any interaction with it is behind a lock. ### Minor refactors included: * Utility methods added to extract configs from the click options * Utility method to add common options to click commands. * All writers moved to using the class approach which extracts a lot of the common code so there's less copy-paste when new runners are added. * Use `@property` for source metadata on base ingest doc to add logic to call `update_source_metadata` if it's still `None` at the time it's fetched. ### Additional bug fixes included * Fsspec connectors were not serializable due to the `ingest_doc_cls`. This was removed from the fields captured by the `@dataclass` decorator and added in a `__post_init__` method. * Various reddit connector params were missing. This doesn't have an explicit ingest test at the moment so was never caught. * Fsspec connector had the parent `update_source_metadata` misnamed as `update_source_metadata_metadata` so it was never being called. ### Flow Diagram ![ingest_pipeline](https://github.com/Unstructured-IO/unstructured/assets/136338424/be485606-cfe0-4931-8b81-c2bf569cf1e2)
2023-10-06 14:49:29 -04:00
processor_config=self.processor_config,
read_config=self.read_config,
record_type=record_type,
record_id=record["Id"],
2023-09-02 08:50:31 -07:00
),
)
except SalesforceMalformedRequest as e:
raise SalesforceMalformedRequest(f"Problem with Salesforce query: {e}")
return ingest_docs