The size of the request headers is too long

[cmilanes93]

• Azure Cosmos: 4.9.0
• Python Version: 3.11.9

Describe the bug
Im encountering an HTTP 400 error indicating that the request headers are too large when using the Azure Cosmos DB Python SDK.

To Reproduce
Steps to reproduce the behavior:

1. import logging
   import time
   import json
   import pandas as pd
   from azure.identity import DefaultAzureCredential
   from azure.cosmos import CosmosClient

logging.basicConfig(level=logging.INFO)

Conexión a Cosmos DB

COSMOS_ENDPOINT = "#####"
DATABASE_NAME = "####"
CONTAINER_NAME = "####"

credential = DefaultAzureCredential()
client = CosmosClient(COSMOS_ENDPOINT, credential=credential)

database = client.get_database_client(DATABASE_NAME)
container = database.get_container_client(CONTAINER_NAME)

Partición (en este caso solo hay una)

PARTITION_KEY_VALUE = "###"

Diccionario para guardar métricas y SQL de cada consulta

metricas_consultas = {}

def ejecutar_consulta(sql_query, parameters=None, partition_key=None,
consulta_nombre="consulta_sin_nombre", descripcion=""):
"""
Ejecuta una consulta en Cosmos DB, mide el tiempo y guarda resultados en metricas_consultas.
"""
start_time = time.time()
# Ejecución de la consulta
results_iter = container.query_items(
query=sql_query,
parameters=parameters or [],
partition_key=partition_key
# NOTA: No se usa enable_cross_partition_query porque se tiene una sola partición.
)
results_list = list(results_iter)
elapsed = time.time() - start_time

# Guardar métricas
metricas_consultas[consulta_nombre] = {
    "pregunta": descripcion,
    "sql_query": sql_query,
    "tiempo_seg": round(elapsed, 4),
    "num_documentos": len(results_list)
}

logging.info(f"[{consulta_nombre}] -> {len(results_list)} resultados en {elapsed:.4f} seg")
return results_list

Query:
consulta_nombre = "group_by_factor_riesgo"
descripcion = "Agrupa (FACTOR_DE_RIESGO, TIPO_RIESGO) sin filtros"

sql_query = """
SELECT
c.I_GRAVEDAD,
c.I_PROBABILIDAD,
COUNT(1)
FROM c
GROUP BY c.I_GRAVEDAD, c.I_PROBABILIDAD
"""

results = ejecutar_consulta(
sql_query=sql_query,
partition_key=PARTITION_KEY_VALUE,
consulta_nombre=consulta_nombre,
descripcion=descripcion
)

for r in results:
print(r)

Expected behavior
Output 16 results.

Screenshots

Additional context
Add any other context about the problem here.

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @AbhinavTrips @bambriz @Pilchie @pjohari-ms @simorenoh.

[Pilchie]

Tagging @kushagraThapar too. Given this is using EntraID, I wonder if the token is too long with claims?

[simorenoh]

Hi @cmilanes93, thank you for using our SDK and opening this issue. This is a known behavior that can happen when a mix of using AAD and queries with long enough continuation tokens. AAD tokens generated from managed/service identities can be quite large depending on the claims on the identity, and the service has a limit of 16KB for request headers - the mix of a large continuation token and a large AAD token can be pushing the request past the limit.

Could you try leveraging the continuation_token_limit request option when running the query and see if it works?

[github-actions]

Hi @cmilanes93. Thank you for opening this issue and giving us the opportunity to assist. To help our team better understand your issue and the details of your scenario please provide a response to the question asked above or the information requested above. This will help us more accurately address your issue.

[lmazuel]

Hi @cmilanes93
We believe this issue has been covered part of #27625. Could you take a look, and let us know if that's not the case? Thanks!