Performance Tuning, Cost Optimization / Internals, Research by Dmitry Tolpeko
Amazon EMR allows you to define scale-out and scale-in rules to automatically add and remove instances based on the metrics you specify.
In this article I am going to explore the instance controller logs that can be very useful in monitoring the auto-scaling. The logs are located in /emr/instance-controller/log/ directory on the EMR master node.
When you run many Hadoop clusters it is useful to automatically collect metrics from all clusters in a single place (Hive table i.e.).
This allows you to perform any advanced and custom analysis of your clusters workload and not be limited to the features provided by Hadoop Administration UI tools that often offer only per cluster view so it is hard to see the whole picture of your data platform.
Snowflake provides various options to monitor data ingestion from external storage such as Amazon S3. In this article I am going to review QUERY_HISTORY and COPY_HISTORY table functions.
The COPY commands are widely used to move data into Snowflake on a time-interval basis, and we can monitor their execution accessing the query history with query_type = 'COPY' filter.
Snowflake separates compute and storage, so it is typical to have a dedicated compute cluster (virtual warehouse) to handle data ingestion into Snowflake (if you do not use Snowpipe).
Like reporting and Ad-hoc SQL, data ingestion has some specifics. Usually there are many tables (data sources) that have own schedule (daily, hourly, every 5, 10, 15 minutes etc.) for periodic data transfer. ETL processes can be overlapped, and can have spikes followed by idle time and so on.
Sometimes you need to reload the entire data set from the source storage into Snowflake. For example, you may want to fully refresh a quite large lookup table (2 GB compressed) without keeping the history. Let’s see how to do this in Snowflake and what issues you need to take into account.
Snowflake uses a cloud storage service such as Amazon S3 as permanent storage for data (Remote Disk in terms of Snowflake), but it can also use Local Disk (SSD) to temporarily cache data used by SQL queries. Let’s test Remote and Local I/O performance by executing a sample SQL query multiple times on X-Large and Medium size Snowflake warehouses:
SELECT MIN(event_hour), MAX(event_hour) FROM events WHERE event_name = 'LOGIN';
Note that you should disable the Result Cache for queries in your session to perform such tests, otherwise Snowflake will just return the cached result immediately after the first attempt:
alter session set USE_CACHED_RESULT = FALSE;
Snowflake stores table data in micro-partitions and uses the columnar storage format keeping MIN/MAX values statistics for each column in every partition and for the entire table as well. Let’s investigate how this affects the performance of SQL queries involving MIN and MAX functions.
Analyzing a Hadoop cluster I noticed that it runs 2 GB and 4 GB containers only, and does not allocate the entire available memory to applications always leaving about 150 GB of free memory.
The clusters run Apache Pig and Hive applications, and the default settings (they are also inherited by Tez engine used by Pig and Hive):
-- from mapred-site.xml mapreduce.map.memory.mb 1408 mapreduce.reduce.memory.mb 2816 yarn.app.mapreduce.am.resource.mb 2816
After creating an Amazon EMR cluster with Spark support, and running a spark application you can notice that the Spark job creates too many tasks to process even a very small data set.
For example, I have a small table country_iso_codes having 249 rows and stored in a comma-delimited text file with the length of 10,657 bytes.
When running the following application on Amazon EMR 5.7 cluster with Spark 2.1.1 with the default settings I can see the large number of partitions generated:
One of our event tables is very large, it contains billions of rows per day. Since the analytics team is often interested in specific events only it makes sense to process the raw events and generate a partition for every event type. Then when a data analyst runs an ad-hoc query, it reads data for the required event type only increasing the query performance.
The problem is that there are 3,000 map tasks are launched to read the daily data and there are 250 distinct event types, so the mappers will produce 3,000 * 250 = 750,000 files per day. That’s too much.