David Ribeiro Alves has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/8823 )

Change subject: KUDU-1704: add c++ client support for READ_YOUR_WRITES mode
......................................................................


Patch Set 10:

(1 comment)

http://gerrit.cloudera.org:8080/#/c/8823/10/src/kudu/client/scan_token-internal.cc
File src/kudu/client/scan_token-internal.cc:

http://gerrit.cloudera.org:8080/#/c/8823/10/src/kudu/client/scan_token-internal.cc@186
PS10, Line 186:     
client->data_->UpdateLatestObservedTimestamp(message.propagated_timestamp());
> Right, so KuduScanner::Open is called after this, and at that point it grab
RYW with scan tokens is a bit of a fuzzy concept, since RYW is by definition 
local and scan tokens are by definition multi client.

If I understand the suggestion, the point is that to get RYW in the point of 
view of the scan token emitter, we could safely ignore the scan token's 
receiver client's internal last propagated timestamp, which might be more 
recent.

I think things get murky here a bit, since RYW with scan tokens are murky. I 
don't think it would necessarily be wrong to do that, but I think it makes 
reasoning about timestamps and how they get propagated a bit harder if we have 
two ways to choose the lower bound for the RYW scan. Moreover, in practice, I 
suspect it's very rarely the case that an active client is permanently being 
written to and read from (using scan tokens).

My suggestion is to keep the path simple and the same as for the local scans. 
Maybe note somewhere that is a possible optimization if we ever notice scan 
token RYW reads stalling.



--
To view, visit http://gerrit.cloudera.org:8080/8823
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I34214245a78aed172a28fbdb395ff5bccd0fc0e1
Gerrit-Change-Number: 8823
Gerrit-PatchSet: 10
Gerrit-Owner: Hao Hao <hao....@cloudera.com>
Gerrit-Reviewer: Alex Rodoni <arod...@cloudera.com>
Gerrit-Reviewer: Dan Burkert <danburk...@apache.org>
Gerrit-Reviewer: David Ribeiro Alves <davidral...@gmail.com>
Gerrit-Reviewer: Hao Hao <hao....@cloudera.com>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Mike Percy <mpe...@apache.org>
Gerrit-Reviewer: Tidy Bot
Gerrit-Comment-Date: Fri, 02 Mar 2018 23:01:02 +0000
Gerrit-HasComments: Yes

Reply via email to