[SOLR-14160] Abstract out the replica identification interface for internode requests

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Noble Paul
• Reporter: Noble Paul
• Created: 03/Jan/20 04:37
• Updated: 21/Jan/20 16:56

Description

The list of replicas to which requests are sent are identified in a static fashion. There is no way to cleanly plugin alternate logic of identifying the replica list .

two major interfaces abstracted out are

 public interface HeaderInterceptor {
  /**Names of headers to be read from the response
   */
  List<String> responseHeaders();

  /**The callback received after a response. This must be called with  a 'null' if there was no such header
   *
   */
  void responseHeaderVal(String key, String val);

  /**The headers that must be added to the request.
   * @return return null or an empty map if nothing needs to added
   */
  Map<String, String > requestHeaders();

}

/**This interface gives urls one by one instead of giving a full list.
 * This also lets the implementation provide extra details as HTTP headers
 *
 */
public interface UrlProvider extends HeaderInterceptor {

  /**Give the base url for the next request to be fired. This is invoked the first time
   * a request needs to be fired, or subsequent after {@link UrlProvider#toRetry()} returns true
   * @return null if no more nodes/cores are available
   */
  String nextBaseurl();


  /**Verify header value. This callback WILL be given if responseHeaderName is supplied, even if it is null
   * @return true if the request should be retried
   * Return the error message if there is something wrong with the response
   */
  boolean toRetry() throws SolrException;

}

Issue Links

• is depended upon by: SOLR-13951 Avoid replica state updates to state.json