[GitHub] [cassandra] jrwest commented on a change in pull request #659: CASSANDRA-15907 Operational Improvements & Hardening for Replica Filtering Protection

[GitBox <gi...@apache.org>]

jrwest commented on a change in pull request #659:
URL: https://github.com/apache/cassandra/pull/659#discussion_r460583875



##########
File path: src/java/org/apache/cassandra/service/DataResolver.java
##########
@@ -150,46 +150,46 @@ private PartitionIterator resolveWithReplicaFilteringProtection()
     {
         // Protecting against inconsistent replica filtering (some replica returning a row that is outdated but that
         // wouldn't be removed by normal reconciliation because up-to-date replica have filtered the up-to-date version
-        // of that row) works in 3 steps:
-        //   1) we read the full response just to collect rows that may be outdated (the ones we got from some
-        //      replica but didn't got any response for other; it could be those other replica have filtered a more
-        //      up-to-date result). In doing so, we do not count any of such "potentially outdated" row towards the
-        //      query limit. This simulate the worst case scenario where all those "potentially outdated" rows are
-        //      indeed outdated, and thus make sure we are guaranteed to read enough results (thanks to short read
-        //      protection).
-        //   2) we query all the replica/rows we need to rule out whether those "potentially outdated" rows are outdated
-        //      or not.
-        //   3) we re-read cached copies of each replica response using the "normal" read path merge with read-repair,
-        //      but where for each replica we use their original response _plus_ the additional rows queried in the
-        //      previous step (and apply the command#rowFilter() on the full result). Since the first phase has
-        //      pessimistically collected enough results for the case where all potentially outdated results are indeed
-        //      outdated, we shouldn't need further short-read protection requests during this phase.
+        // of that row) involves 3 main elements:
+        //   1) We combine short-read protection and a merge listener that identifies potentially "out-of-date"
+        //      rows to create an iterator that is guaranteed to produce enough valid row results to satisfy the query 
+        //      limit if enough actually exist. A row is considered out-of-date if its merged form is non-empty and we 

Review comment:
       minor typo: "form" instead of "from"




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: pr-unsubscribe@cassandra.apache.org
For additional commands, e-mail: pr-help@cassandra.apache.org