[COUCHDB-3034] CouchDB 2.0: Race condition when creating/deleting database

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Won't Fix
• Affects Version/s: 2.0.0
• Assignee: Unassigned
• Reporter: Will Holley
• Created: 09/Jun/16 13:15
• Updated: 02/Sep/18 07:07
• Resolved: 02/Sep/18 07:07

Description

I've noticed that CouchDB 2.0 (master) occasionally returns 500 errors when multiple clients attempt to PUT or DELETE the same database concurrently. This is reproducible by running multiple instances of following script in parallel against an n=1 (single node) cluster:

#!/bin/bash

for i in `seq 1 2000`;
do
	printf '\nPUT: '
    curl -XPUT 'http://localhost:15984/testdb' -w "%{http_code}\\n"
    printf '\nDELETE: '
    curl -XDELETE 'http://localhost:15984/testdb' -w "%{http_code}\\n"
done

This generates output along the lines of:

PUT: {"ok":true}
201

DELETE: {"ok":true}
200

PUT: {"error":"error","reason":"internal_server_error"}
500

DELETE: {"error":"not_found","reason":"Database does not exist."}
404

PUT: {"error":"file_exists","reason":"The database could not be created, the file already exists."}
412

DELETE: {"ok":true}
200

PUT: {"error":"file_exists","reason":"The database could not be created, the file already exists."}
412

DELETE: {"ok":true}
200

PUT: {"ok":true}
201

DELETE: {"ok":true}
200

I can't reproduce the error against CouchDB 1.6 or CouchDB master on port 15986 (bypassing the clustering).