Database Reference
In-Depth Information
Here's the answer that I got from Jesse Yates. Keep in mind that this was true at the
time of writing this topic and the situation will most likely be further along at the
time of your reading the topic.
As Jesse said, "I worked on adding tracing into Phoenix, but never finished it up; it
would use Cloudera's HTrace library since that is already bundled into HBase 0.96
and higher versions. With the 4.0 Phoenix branch (which is based on HBase 0.96
and higher versions), this integration should be even easier, and might actually
get inished!"
Until then, you can explain this as follows:
The URL https://github.com/jyates/phoenix/tree/tracing worked at the
time of Hadoop 2, but never quite brought it the extra mile necessary to roll it into
Phoenix properly. Interestingly, it has also been the main reason why I made the
build multimodule (and included the Hadoop Compatibility stuff).
So, there is little doubt that tracing is available. We can also expect it to be available
with the Java API, and this is how you can approach measuring performance in
production, by writing a custom load testing code.
Please keep in mind that tracing has its own overhead, so use it only for debugging
and profiling and not for overall performance measurements.
As an advanced exercise, solve this:
What will you use to fix the performance problem of having too many maps?
Hint
Use another table to index locations. Share this solution on the
topic's forum.
A popularity contest
Imagine you need to record the popularity of videos. If someone gives a rating,
this increases the total number of ratings and the exact rating that the user gave
goes towards calculating the average rating of the video. How will you design
a table for this?
Please close the topic and try to design your solution.
 
Search WWH ::




Custom Search