Elasticsearch Bulk insert w/ Python - socket timeout error

Question

ElasticSearch 7.10.2

Python 3.8.5

elasticsearch-py 7.12.1

I'm trying to do a bulk insert of 100,000 records to ElasticSearch using elasticsearch-py bulk helper.

Here is the Python code:

import sys
import datetime
import json
import os
import logging
from elasticsearch import Elasticsearch
from elasticsearch.helpers import streaming_bulk

     # ES Configuration start
        es_hosts = [
        "http://localhost:9200",]
        es_api_user = 'user'
        es_api_password = 'pw'
        index_name = 'index1'
        chunk_size = 10000
        errors_before_interrupt = 5
        refresh_index_after_insert = False
        max_insert_retries = 3
        yield_ok = False  # if set to False will skip successful documents in the output
    
        # ES Configuration end
        # =======================
    
        filename = file.json
    
        logging.info('Importing data from {}'.format(filename))
    
        es = Elasticsearch(
            es_hosts,
            #http_auth=(es_api_user, es_api_password),
            sniff_on_start=True,  # sniff before doing anything
            sniff_on_connection_fail=True,  # refresh nodes after a node fails to respond
            sniffer_timeout=60,  # and also every 60 seconds
            retry_on_timeout=True,  # should timeout trigger a retry on different node?
        )
    
    
        def data_generator():
            f = open(filename)
            for line in f:
                yield {**json.loads(line), **{
                    "_index": index_name,
                }}
    
    
        errors_count = 0
    
        for ok, result in streaming_bulk(es, data_generator(), chunk_size=chunk_size, refresh=refresh_index_after_insert,
                                         max_retries=max_insert_retries, yield_ok=yield_ok):
            if ok is not True:
                logging.error('Failed to import data')
                logging.error(str(result))
                errors_count += 1
    
                if errors_count == errors_before_interrupt:
                    logging.fatal('Too many import errors, exiting with error code')
                    exit(1)
                    
        print("Documents loaded to Elasticsearch")

When the json file contains a small amount of documents (~100), this code runs without issue. But I just tested it with a file of 100k documents, and I got this error:

WARNING:elasticsearch:POST http://127.0.0.1:9200/_bulk?refresh=false [status:N/A request:10.010s]
Traceback (most recent call last):
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/connectionpool.py", line 426, in _make_request
    six.raise_from(e, None)
  File "<string>", line 3, in raise_from
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/connectionpool.py", line 421, in _make_request
    httplib_response = conn.getresponse()
  File "/Users/me/opt/anaconda3/lib/python3.8/http/client.py", line 1347, in getresponse
    response.begin()
  File "/Users/me/opt/anaconda3/lib/python3.8/http/client.py", line 307, in begin
    version, status, reason = self._read_status()
  File "/Users/me/opt/anaconda3/lib/python3.8/http/client.py", line 268, in _read_status
    line = str(self.fp.readline(_MAXLINE + 1), "iso-8859-1")
  File "/Users/me/opt/anaconda3/lib/python3.8/socket.py", line 669, in readinto
    return self._sock.recv_into(b)
socket.timeout: timed out

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/elasticsearch/connection/http_urllib3.py", line 251, in perform_request
    response = self.pool.urlopen(
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/connectionpool.py", line 726, in urlopen
    retries = retries.increment(
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/util/retry.py", line 386, in increment
    raise six.reraise(type(error), error, _stacktrace)
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/packages/six.py", line 735, in reraise
    raise value
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/connectionpool.py", line 670, in urlopen
    httplib_response = self._make_request(
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/connectionpool.py", line 428, in _make_request
    self._raise_timeout(err=e, url=url, timeout_value=read_timeout)
  File "/Users/me/opt/anaconda3/lib/python3.8/site-packages/urllib3/connectionpool.py", line 335, in _raise_timeout
    raise ReadTimeoutError(
urllib3.exceptions.ReadTimeoutError: HTTPConnectionPool(host='127.0.0.1', port=9200): Read timed out. (read timeout=10)

I have to admit this one is a bit over my head. I don't typically like to paste large error messages here, but I'm not sure what about this message is relevant.

I can't help but think that I maybe need to adjust some of the params in the es object? Or the configuration variables? I don't know enough about the params to be able to make an educated decision on my own.

And the last but certainly not least point - it looks like some documents were loaded into the ES index nonetheless. But even stranger, the count shows 110k when the json file only has 100k.

Richard E · Accepted Answer · 2022-10-31 12:24:43Z

12

+50

TL;DR:

Reduce the chunk_size from 10000 to the default of 500 and I'd expect it to work. You probably want to disable automatic retries if that can give you duplicates.

What happened?

When creating your Elasticsearch object, you specified chunk_size=10000. This means that the streaming_bulk call will try to insert chunks of 10000 elements. The connection to elasticsearch has a configurable timeout, which by default is 10 seconds. So, if your elasticsearch server takes more than 10 seconds to process the 10000 elements you want to insert, a timeout will happen and this will be handled as an error.

When creating your Elasticsearch object, you also specified retry_on_timeout as True and in the streaming_bulk_call you set max_retries=max_insert_retries, which is 3.

This means that when such a timeout happens, the library will try reconnecting 3 times, however, when the insert still has a timeout after that, it will give you the error you noticed. (Documentation)

Also, when the timeout happens, the library can not know whether the documents were inserted successfully, so it has to assume that they were not. Thus, it will try to insert the same documents again. I don't know how your input lines look like, but if they do not contain an _id field, this would create duplicates in your index. You probably want to prevent this -- either by adding some kind of _id, or by disabling the automatic retry and handling it manually.

What to do?

There is two ways you can go about this:

Increase the timeout
Reduce the chunk_size

streaming_bulk by default has chunk_size set to 500. Your 10000 is much higher. I wouldn't expect a high performance gain when increasing this to more than 500, so I'd advice you to just use the default of 500 here. If 500 still fails with a timeout, you may even want to reduce it further. This could happen if the documents you want to index are very complex.

You could also increase the timeout for the streaming_bulk call, or, alternatively, for your es object. To only change it for the streaming_bulk call, you can provide the request_timeout keyword argument:

for ok, result in streaming_bulk(
        es,
        data_generator(),
        chunk_size=chunk_size,
        refresh=refresh_index_after_insert,
        request_timeout=60*3,  # 3 minutes
        yield_ok=yield_ok):
    # handle like you did
    pass

However, this also means that elasticsearch node failure will only be detected after this higher timeout. See the documentation for more details

edited Oct 31, 2022 at 12:24

answered May 18, 2021 at 11:27

Richard E

3,4773 gold badges19 silver badges37 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Stpete111 Over a year ago

Thanks for the well thought out answer. I reset the chunk size to the default of 500. This fixed the issue. It did create a significant impact on performance (20 seconds to load 100k docs using chunk size of 10000 vs 1.5 minutes using chunk size of 500). That's certainly to be expected, and this post is about addressing the errors, not performance. I'll play with chunk size until I find the happy medium between performance and not getting errors.

Richard E Over a year ago

That is very interesting, I did not expect that. In this case, it may also make sense to keep a higher chunk_size and instead increase the timeout.

Richard E Over a year ago

I have updated my answer, it now contains an example call with custom timeout for only the streaming_bulk call. The key element is the kwarg request_timeout.

Collectives™ on Stack Overflow

Elasticsearch Bulk insert w/ Python - socket timeout error

1 Answer 1

TL;DR:

What happened?

What to do?

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

TL;DR:

What happened?

What to do?

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related