[ https://issues.apache.org/jira/browse/TS-4701?focusedWorklogId=30244&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-30244 ]
ASF GitHub Bot logged work on TS-4701: -------------------------------------- Author: ASF GitHub Bot Created on: 06/Oct/16 21:22 Start Date: 06/Oct/16 21:22 Worklog Time Spent: 10m Work Description: Github user PSUdaemon commented on a diff in the pull request: https://github.com/apache/trafficserver/pull/1084#discussion_r82288142 --- Diff: proxy/ParentSelection.h --- @@ -59,12 +59,7 @@ enum ParentResultType { PARENT_FAIL, }; -enum ParentRR_t { - P_NO_ROUND_ROBIN = 0, - P_STRICT_ROUND_ROBIN, - P_HASH_ROUND_ROBIN, - P_CONSISTENT_HASH, -}; +enum ParentRR_t { P_NO_ROUND_ROBIN = 0, P_STRICT_ROUND_ROBIN, P_HASH_ROUND_ROBIN, P_CONSISTENT_HASH, P_LATCHED_ROUND_ROBIN }; --- End diff -- Ok, I guess that is just an unfortunate name for that enum. Issue Time Tracking ------------------- Worklog Id: (was: 30244) Time Spent: 1h 10m (was: 1h) > Add a new latched parent selection strategy. > -------------------------------------------- > > Key: TS-4701 > URL: https://issues.apache.org/jira/browse/TS-4701 > Project: Traffic Server > Issue Type: New Feature > Components: Parent Proxy > Reporter: John Rushford > Assignee: John Rushford > Fix For: sometime > > Time Spent: 1h 10m > Remaining Estimate: 0h > > For live linear video we are utilizing parent selection to round robin to > origin servers in the event of a origin server failure, parent_is_proxy=false > and round_robin=false. With this round robin strategy, the first parent in > the parent.config list is always selected. In the event of a connection > failure, a new parent is selected from the list. This works fine except that > switching to a new parent is expensive, in that the client has to retune > because the abr manifests between origins are different. A single retune can > be tolerated but, after the parent retry time has elapsed, the round robin > strategy will revert back to the original parent when it becomes available. > This causes a second retune. In order to minimize client retuning, I'd like > to add a new round robin strategy that is similar to round_robin=false but > the new strategy would stay "latched" to the selected parent and remain > "latched" until such time that a connection failure warrants using a new > parent. I propose calling this strategy "latched" as in an electronic latch > that becomes fixed in a particular state when triggered. This strategy would > be a minor modification to ParentRoundRobin and ParentSelection. -- This message was sent by Atlassian JIRA (v6.3.4#6332)