Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
If you're worried about it then accept the alternative PR I linked.
On Sun, Oct 9, 2016 at 11:37 PM, Shixiong Zhu
wrote:
> During the original implementation I had verified th
Github user zsxwing commented on the issue:
https://github.com/apache/spark/pull/15387
> During the original implementation I had verified that calling pause
kills the internal message buffer, which is one of the complications leading to
a cached consumer per partition.
I obs
Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
Let me know if you guys like that alternative PR better
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not hav
Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
If the concern is TD's comment,
"Future calls to {@link #poll(long)} will not return any records from these
partitions until they have been resumed using {@link #resume(Collection)}."
http
Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
Poll also isn't going to return you just messages for a single
topicpartition, so to do what you're suggesting you'd have to go through
all the messages and do additional processing, not jus
Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
You dont want poll consuming messages, its not just about offset
correctness, the driver shouldnt be spending time or bandwidth doing that.
What is the substantive concern with this solution
Github user zsxwing commented on the issue:
https://github.com/apache/spark/pull/15387
#15397 is the fix for structured streaming.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user zsxwing commented on the issue:
https://github.com/apache/spark/pull/15387
@koeninger If `poll(0)` returns a non empty results, how about just
subtract the offsets back using the result size? Then we don't need to count on
`poll(0)` not updating the offsets.
---
If your
Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
I set auto commit to false, and still recreated the test failure.
That makes sense to me, consumer position should still be getting updated
in memory even if it isn't saved to storage a
Github user zsxwing commented on the issue:
https://github.com/apache/spark/pull/15387
> but polling ordinarily consumes messages and adjusts position.
Even if `enable.auto.commit` is `false`? In the doc, it says `automatically
set as the last committed offset`, so I guess set
Github user koeninger commented on the issue:
https://github.com/apache/spark/pull/15387
I'm not going to say anything is impossible, which is the point of the
assert. If it does somehow happen, it will be at start, so should be
obvious.
The whole poll 0 / pause thing i
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/15387
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
e
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/15387
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/66479/
Test PASSed.
---
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/15387
**[Test build #66479 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/66479/consoleFull)**
for PR 15387 at commit
[`aca55de`](https://github.com/apache/spark/commit/
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/15387
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
e
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/15387
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/66477/
Test PASSed.
---
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/15387
**[Test build #66477 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/66477/consoleFull)**
for PR 15387 at commit
[`1fc5863`](https://github.com/apache/spark/commit/
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/15387
**[Test build #66479 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/66479/consoleFull)**
for PR 15387 at commit
[`aca55de`](https://github.com/apache/spark/commit/a
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/15387
**[Test build #66477 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/66477/consoleFull)**
for PR 15387 at commit
[`1fc5863`](https://github.com/apache/spark/commit/1
19 matches
Mail list logo