> On Nov. 20, 2014, 5:09 p.m., Maxim Khutornenko wrote:
> > src/main/java/org/apache/aurora/scheduler/storage/log/EntrySerializer.java,
> > line 92
> > <https://reviews.apache.org/r/28306/diff/1/?file=771713#file771713line92>
> >
> > Mind commenting on the algorithm here? Why the magic "-2"?
>
> Kevin Sweeney wrote:
> Not sure what I'd say there that doesn't risk contradicting the code
> below - we don't usually comment loop variables and the old version had the
> same amount of commenting.
>
> Maxim Khutornenko wrote:
> I just find it harder to follow when I see anything less than -1 as the
> initial condition. Why not starting with -1 (to account for the header) and
> then increment i at the end of computeNext()?
Each branch short-circuits so I'd have to have a result intermediate value. I
don't find that more readable
```java
byte[] result;
if (i == -1) {
result = header;
} else if (...) {
result = encode(...);
} else {
return endOfFile();
}
i++;
return result;
```
- Kevin
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/28306/#review62481
-----------------------------------------------------------
On Nov. 20, 2014, 5:10 p.m., Kevin Sweeney wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/28306/
> -----------------------------------------------------------
>
> (Updated Nov. 20, 2014, 5:10 p.m.)
>
>
> Review request for Aurora, David McLaughlin and Bill Farner.
>
>
> Bugs: AURORA-930
> https://issues.apache.org/jira/browse/AURORA-930
>
>
> Repository: aurora
>
>
> Description
> -------
>
> Stream frame chunks instead of preallocating them.
>
> This avoids allocating an entire additional snapshot worth of heap during
> entry serialization, reducing overall heap impact of the serializer from
> 2*sizeof(serialized-entry) to sizeof(serialized-entry)+chunkSize.
>
> Further optimizations out-of-scope for this change:
>
> * Make the returned iterator mutate a fixed-size buffer (for GC pressure
> avoidance).
> * Change the log format so that FrameHeader doesn't need to know the size and
> checksum of the serialized data ahead-of-time (maybe write it as a trailer).
>
>
> Diffs
> -----
>
> src/main/java/org/apache/aurora/scheduler/storage/log/EntrySerializer.java
> f4fa1cb740633ced529c1b5fd9f18abba8944571
>
> src/main/java/org/apache/aurora/scheduler/storage/log/StreamManagerImpl.java
> cb95d8996a934751745f423b79279266d73b7722
> src/test/java/org/apache/aurora/scheduler/app/SchedulerIT.java
> c90389433d81dd72756c659736e38fd9f66fcb35
>
> Diff: https://reviews.apache.org/r/28306/diff/
>
>
> Testing
> -------
>
> ./gradlew -Pq build
>
>
> Thanks,
>
> Kevin Sweeney
>
>
