Kensho Hirasawa created SOLR-11107:
--------------------------------------
Summary: make the threshold for zombie servers of LBHttpSolrClient
configurable
Key: SOLR-11107
URL: https://issues.apache.org/jira/browse/SOLR-11107
Project: Solr
Issue Type: Improvement
Security Level: Public (Default Security Level. Issues are Public)
Components: clients - java
Reporter: Kensho Hirasawa
LBHttpSolrClient treats a server as a zombie if it receives abnormal response
from the server. Servers in bad status are detected by this scheme.
However, healthy servers can be treated as a zombie because even healthy
servers sometimes returns server error. For example, if we submit a range query
with bad syntax (like [a TO 1]), servers return 500 server error and the
servers are treated as zombies. To make matters worse, in SolrCloud mode, all
the servers in a cluster become zombies by retry if we do not set
numServersToTry. Even if we set some small number to numServersToTry, several
evil requests can rob the client of load balancing function.
Servers should return 400 when they receive bad request but it may be difficult
to fix all. Therefore, it is useful to enable relaxing the threshold for
treating a server as a zombie. The implemented solution judges zombies by how
many times in a row a server returns bad response not by whether a server
returns bad response even once.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
