AWS (Amazon Redshift) presentation

Editor's Notes

• #4: Amazon Redshift data warehouse is a cloud based massively parallel processing (MPP), columnar database that consists of multiple computers (nodes). It delivers fast query performance by using columnar storage technology to improve I/O efficiency and parallelizing queries across multiple nodes. Redshift uses standard PostgreSQL ODBC drivers, allowing the usage of wide range of familiar SQL clients. Most common tasks associated with provisioning, configuring and monitoring a data warehouse are automated within Redshift for the ease of administration.
• #6: Dense Storage Node Types Node Size vCPU ECU RAM (GiB) Slices Per Node Storage Per Node Node Range Total Capacity ds1.xlarge 2 4.4 15 2 2 TB HDD 1–32 64 TB ds1.8xlarge 16 35 120 16 16 TB HDD 2–128 2 PB ds2.xlarge 4 13 31 2 2 TB HDD 1–32 64 TB ds2.8xlarge 36 119 244 16 16 TB HDD 2–128 2 PB Dense Compute Node Types Node Size vCPU ECU RAM (GiB) Slices Per Node Storage Per Node Node Range Total Capacity dc1.large 2 7 15 2 160 GB SSD 1–32 5.12 TB dc1.8xlarge 32 104 244 32 2.56 TB SSD 2–128 326 TB Cluster architecture Clients connect via existing protocols to the Leader Node. Leader node develops a query plan and may generate and compile C++ code to be executed by the compute nodes Leader node will distribute work across compute nodes using Distribution Keys (more later) Compute nodes receive work from leader node and may transmit data amongst themselves to answer the query Leader aggregates the results and returns to client Leader can distribute bulk data loads across compute nodes http://docs.aws.amazon.com/redshift/latest/dg/c_high_level_system_architecture.html
• #9: For example, suppose a table contains 100 columns. A query that uses five columns will only need to read about five percent of the data contained in the table. This savings is repeated for possibly billions or even trillions of records for large databases. In contrast, a row-wise database would read the blocks that contain the 95 unneeded columns as well.Typical database block sizes range from 2 KB to 32 KB. Amazon Redshift uses a block size of 1 MB, which is more efficient and further reduces the number of I/O requests needed to perform any database loading or other operations that are part of query execution. Недолік Пысля апдейта ынсерта або деліта потрібно ранити вакууь і аналайз Тяжке для операцій які модифікують дані
• #11: Compression is a column-level operation that reduces the size of data when it is stored. Compression conserves storage space and reduces the size of data that is read from storage, which reduces the amount of disk I/O and therefore improves query performance. By default, Amazon Redshift stores data in its raw, uncompressed format. When you create tables in an Amazon Redshift database, you can define a compression type, or encoding, for the columns.  For us, the most expensive part of Redshift is the data storage. Thus, the more we can compress the data to save space, the better. We can typically compress our data about 3x in Redshift. This means that Redshift’s cost of $1000/TB/year (for 3 year reserved nodes) is actually more like $350 per uncompressed TB per year. Redshift allows you to specify a different compression type for each column, which is great. Instead of using a generic strategy to compress all of your data, you should use the compression type that best suits the data in the column. In addition to compression, if you know you’ll be using Redshift on a continuous, long-term basis, another way to save costs is to use reserved instances. By using reserved instances, we save about 60% of the cost compared to on-demand Redshift. The syntax is as follows: CREATE TABLE table_name (column_name data_type ENCODE encoding-type)[, ...] create table product( product_id int, product_name char(20) encode bytedict); You cannot change the compression encoding for a column after the table is created.
• #12: Додати формати даних які підтримуються у редшіфті
• #13: Use SORTKEY SORTKEY essentially defines how the data will be sorted in the storage. This feature is useful to limit the amount of data that has to be scanned. For example, if I have a large table full of news paper articles over a century and want to find article published between 1980 - 1985 that mention "Tiger", it's useful to have articles sorted by published_date on the storage, because that way I can limit the scanning on blocks that contain these dates. They are also useful for joining if the key is also the DISTKEY because the query planner can skip a lot of work. You *can* specify multiple SORTKEYs. When you specify SORTKEY(a, b), the data is effectively sorted as if with "ORDER BY (a, b). If cardinality of a is high enough, filtering by a is very effective, but having a second SORTKEY will make small sense, and vice versa. Therefore the utility of setting multiple SORTKEY is more difficult to judge. Start with a single SORTKEY and see how it goes. After data loads or inserts, the ANALYZE command should be run: ANALYZE updates the table metadata that is used by the query planner – very important for column-based data storage and ongoing query performance
• #16: Дані не сортовані по регіону і кантрі
• #17: Дані не сортовані по регіону і кантрі Sort Keys – Comparing StylesIncreased load and vacuum times More effective with large tables (> 100M+ rows) Use Compound Sort Key when appending data in order Sort Keys – Interleaved Considerations
• #18: Додати опис по скев SkewDistribution Styles When you create a table, you designate one of three distribution styles: KEY, ALL, or EVEN. KEY distribution The rows are distributed according to the values in one column. The leader node will attempt to place matching values on the same node slice. If you distribute a pair of tables on the joining keys, the leader node collocates the rows on the slices according to the values in the joining columns so that matching values from the common columns are physically stored together. ALL distribution A copy of the entire table is distributed to every node. Where EVEN distribution or KEY distribution place only a portion of a table's rows on each node, ALL distribution ensures that every row is collocated for every join that the table participates in. EVEN distribution The rows are distributed across the slices in a round-robin fashion, regardless of the values in any particular column. EVEN distribution is appropriate when a table does not participate in joins or when there is not a clear choice between KEY distribution and ALL distribution. EVEN distribution is the default distribution style. Consider choosing DISTKEY What is "DISTKEY" anyways? DISTKEY essentially decides which row goes to which node. For example, if you declare "user_id" as DISTKEY, RedShift will donode_id = hash(user_id) % num_nodes to choose the node to store that row. Well, it's not THAT simple, but you get the idea. Why does it matter? DISTKEY primarily matters when you do a join. Let's say a SQL statement SELECT * FROM User INNER JOIN Post ON (User.UserId = Post.UserId) WHERE Post.Type = 1 is issued. If User and Post both used UserId as DISTKEY, a RedShift node can just take the allocated shard, join them, filter them and send the (much smaller) contribution over the wire to be combined. However, if User was distributed by UserId and Post was distributed byArticleId, Posts that belong to Users on a node will be on other nodes. Therefore the nodes have to ship the entire shard over the network to perform the join, which is expensive. What should I do? If a table is large and you anticipate a join with another large table, then consider choosing the key that will be used for the join to be the DISTKEY. In other words, unless this is the case don't declare a DISTKEY (RedShift will distribute the rows evenly) What is "data skew"? Data skew is when data concentrates on small number of nodes due to a badly chosen DISTKEY. Imagine you have a huge user base which are predominantly located in US. If you use "country_code" as DISTKEY, most of the data will end up on one node because most users will have the same country code "US". This means that this one node will do most of the work while other nodes will remain idle, which is inefficient. Therefore, it's important to choose a DISTKEY that will result in an even(-ish) distribution among the nodes.
• #20: http://docs.aws.amazon.com/redshift/latest/dg/r_SET.html – set label for query group http://docs.aws.amazon.com/redshift/latest/dg/cm-c-defining-query-queues.html#wlm-wildcards Workload Management (WLM) is necessary to optimize access to database resources for concurrently executing queries. The goals of a functional workload management are to Optimally leverage available (hardware) resources for performance and throughput - Prioritize access for high priority jobs Assure resource availability by avoiding system lock-up by any small set of jobs Effective workload management starts with comprehensive monitoring, allowing identifying bottleneck conditions, and then leveraging the available platform tools to implement a workload management strategy. Amazon Redshift’s workload management (WLM) helps you allocate resources to certain user groups or query groups. By adjusting your WLM queue configurations, you can drastically improve performance and query speed. For our Redshift clusters, we use WLM to set what percentage of memory goes to a customer’s queries, versus loading data and other maintenance tasks. If we give a lot of memory to our customers and don’t leave much for loading new data, loading will never finish; if we do the opposite, customer queries will never finish. There has to be a good balance when allocating resources to our customers relative to internal functions. WLM also allows us to define how many queries can be run in parallel at the same time for a given group. If the maximum number of queries are already running in parallel, any new queries will enter a queue to wait for the first one to finish. Of course, the optimal configuration for this will depend on your specific use case, but it’s a tool that anyone using Amazon Redshift should definitely take advantage of. https://www.flydata.com/blog/optimal-wlm-settings-redshift/ Use WLM to counter resource hogging When queries are issued concurrently, resource hogging can become a problem. For example, if somebody issues 10 queries that take 1 hour each, another guy with a 5 min query can wait for a long time before he can get his query done. To prevent this kind of problem, consider using WLM. Work Load Management (WLM)- It enables to create multiple query queues according to different user groups or query groups. This will help in managing workloads.- Redshift allocates an equal, fixed share of server memory to each queue, and, by default, an equal, fixed share of a queue's memory to each query slot in the queue regardless of the number of queries that are actually running concurrently..- WLM assigns a query to a queue based on the user's group or query group specified while running the SQL.Maxmimum concurrency level per queue is 50.Total concurrency of all the queues is 50.For each queue, you can specify• Concurrency level• User groups• Query groups• Wildcards• WLM memory percent to use - To specify the amount of available memory that is allocated to a query - If you specify a WLM Memory Percent to Use value for one or more queues but not all of the queues, then the unallocated memory is divided evenly among the remaining queues.• WLM timeout - To limit the amount of time that queries in a given WLM queue are permitted to use. Specified in milliseconds. - statement timeout configuration parameter applies to the entire cluster, WLM timeoutis specific to a single queue in the WLM configurationIf you have multiple queries that each access data on a single slice, set up a separate WLM queue to execute those queries concurrently. Redshift will assign concurrent queries to separate slices.By increasing concurrency, you increase the contention for system resources and limit the overall throughput.If a specific query needs more memory than is allocated to a single query slot, you can increase the available memory by increasing the wlm_query_slot_count (p. 697) parameterDefault queueThe last queue defined in the WLM configuration is the default queue.You can set the concurrency level and the timeout for the default queue, but it cannot include user groups or query groups. The default queue counts against the limit of eight query queues and the limit of 50 concurrent queries.Superuser queueTo run a query in the Superuser queue, a user must be logged in as a superuser and must run the query within the predefined 'superuser' query group.The WLM configuration is an editable parameter (wlm_json_configuration) in a parameter group, which can be associated with one or more clusters.You must reboot the cluster after changing the WLM configuration.
• #22: To load data from files located in one or more S3 buckets, use the FROM clause to indicate how COPY will locate the files in Amazon S3. You can provide the object path to the data files as part of the FROM clause, or you can provide the location of a manifest file that contains a list of Amazon S3 object paths. The COPY command leverages the Amazon Redshift massively parallel processing (MPP) architecture to read and load data in parallel from files in an Amazon S3 bucket. You can take maximum advantage of parallel processing by splitting your data into multiple files and by setting distribution keys on your tables.  Note We strongly recommend that you divide your data into multiple files to take advantage of parallel processing.
• #23: The COPY command leverages the Amazon Redshift massively parallel processing (MPP) architecture to read and load data in parallel from files on Amazon S3, from a DynamoDB table, or from text output from one or more remote hosts. Loads data into a table from data files or from an Amazon DynamoDB table. The files can be located in an Amazon Simple Storage Service (Amazon S3) bucket, an Amazon Elastic MapReduce (Amazon EMR) cluster, or a remote host that is accessed using a Secure Shell (SSH) connection. The maximum size of a single input row from any source is 4 MB. To use the COPY command, you must have INSERT privilege for the Amazon Redshift table. We recommend using role-based access control because it is provides more secure, fine-grained control of access to AWS resources and sensitive user data, in addition to safeguarding your AWS credentials.To use role-based access control, you must first create an IAM role using the Amazon Redshift service role type, and then attach the role to your cluster. When you create an IAM role, IAM returns an Amazon Resource Name (ARN) for the role. To execute a COPY command using an IAM role, provide the role ARN in the CREDENTIALS parameter string. The following COPY command example uses the role MyRedshiftRole for authentication. copy customer from 's3://mybucket/mydata' credentials 'aws_iam_role=arn:aws:iam::12345678901:role/MyRedshiftRole';
• #24: Step 6: Vacuum and Analyze the Database Whenever you add, delete, or modify a significant number of rows, you should run a VACUUM command and then an ANALYZE command. A vacuum recovers the space from deleted rows and restores the sort order. The ANALYZE command updates the statistics metadata, which enables the query optimizer to generate more accurate query plans. For more information, see Vacuuming Tables. If you load the data in sort key order, a vacuum is fast. In this tutorial, you added a significant number of rows, but you added them to empty tables. That being the case, there is no need to resort, and you didn't delete any rows. COPY automatically updates statistics after loading an empty table, so your statistics should be up-to-date. However, as a matter of good housekeeping, you will complete this tutorial by vacuuming and analyzing your database.
• #25: To clean up tables after a bulk delete, a load, or a series of incremental updates, you need to run the VACUUMcommand, either against the entire database or against individual tables. Sort Stage and Merge Stage Amazon Redshift performs a vacuum operation in two stages: first, it sorts the rows in the unsorted region, then, if necessary, it merges the newly sorted rows at the end of the table with the existing rows. When vacuuming a large table, the vacuum operation proceeds in a series of steps consisting of incremental sorts followed by merges. If the operation fails or if Amazon Redshift goes off line during the vacuum, the partially vacuumed table or database will be in a consistent state, but you will need to manually restart the vacuum operation. Incremental sorts are lost, but merged rows that were committed before the failure do not need to be vacuumed again. If the unsorted region is large, the lost time might be significant. For more information about the sort and merge stages, see Managing the Volume of Merged Rows. Users can access tables while they are being vacuumed. You can perform queries and write operations while a table is being vacuumed, but when DML and a vacuum run concurrently, both might take longer. If you execute UPDATE and DELETE statements during a vacuum, system performance might be reduced. Incremental merges temporarily block concurrent UPDATE and DELETE operations, and UPDATE and DELETE operations in turn temporarily block incremental merge steps on the affected tables. DDL operations, such as ALTER TABLE, are blocked until the vacuum operation finishes with the table. Vacuum Threshold By default, VACUUM skips the sort phase for any table where more than 95 percent of the table's rows are already sorted. Skipping the sort phase can significantly improve VACUUM performance. In addition, in the delete phase VACUUM reclaims space such that at least 95 percent of the remaining rows are not marked for deletion. Because VACUUM can often skip rewriting many blocks that contain only a few rows marked for deletion, it usually needs much less time for the delete phase compared to reclaiming 100 percent of deleted rows. To change the default sort threshold for a single table, include the table name and the TO threshold PERCENT parameter when you run the VACUUM command. Vacuum Types You can run a full vacuum, a delete only vacuum, a sort only vacuum, or a reindex with full vacuum. VACUUM FULL We recommend a full vacuum for most applications where reclaiming space and resorting rows are equally important. It's more efficient to run a full vacuum than to run back-to-back DELETE ONLY and SORT ONLY vacuum operations. VACUUM FULL is the same as VACUUM. Full vacuum is the default vacuum operation. VACUUM DELETE ONLY A DELETE ONLY vacuum is the same as a full vacuum except that it skips the sort. A DELETE ONLY vacuum saves time when reclaiming disk space is important but resorting new rows is not. For example, you might perform a DELETE ONLY vacuum operation if you don't need to resort rows to optimize query performance. VACUUM SORT ONLY A SORT ONLY vacuum saves some time by not reclaiming disk space, but in most cases there is little benefit compared to a full vacuum. VACUUM REINDEX Use VACUUM REINDEX for tables that use interleaved sort keys. REINDEX reanalyzes the distribution of the values in the table's sort key columns, then performs a full VACUUM operation. VACUUM REINDEX takes significantly longer than VACUUM FULL because it needs to take an extra analysis pass over the data, and because merging in new interleaved data can involve touching all the data blocks. If a VACUUM REINDEX operation terminates before it completes, the next VACUUM resumes the reindex operation before performing the vacuum. Examples Reclaim space and database and resort rows in alls tables based on the default 95 percent vacuum threshold. vacuum; Reclaim space and resort rows in the SALES table based on the default 95 percent threshold. vacuum sales; Always reclaim space and resort rows in the SALES table. vacuum sales to 100 percent; Resort rows in the SALES table only if fewer than 75 percent of rows are already sorted. vacuum sort only sales to 75 percent; Reclaim space in the SALES table such that at least 75 percent of the remaining rows are not marked for deletion following the vacuum. vacuum delete only sales to 75 percent; Reindex and then vacuum the LISTING table. vacuum reindex listing; The following command returns an error. vacuum reindex listing to 75 percent;
• #26: Snapshots are point-in-time backups of a cluster. There are two types of snapshots:automated and manual. Amazon Redshift stores these snapshots internally in Amazon S3 by using an encrypted Secure Sockets Layer (SSL) connection. If you need to restore from a snapshot, Amazon Redshift creates a new cluster and imports data from the snapshot that you specify.
• #27: Amazon Virtual Private Cloud (Amazon VPC) enables you to launch Amazon Web Services (AWS) resources into a virtual network that you've defined. This virtual network closely resembles a traditional network that you'd operate in your own data center, with the benefits of using the scalable infrastructure of AWS. When you provision an Amazon Redshift cluster, it is locked down by default so nobody has access to it. To grant other users inbound access to an Amazon Redshift cluster, you associate the cluster with a security group.  Amazon Redshift has multiple levels of security for node, connection, and data security. The following diagram illustrates a number of the capabilities available for security using Redshift. All nodes of a Redshift cluster are contained in an Internal VPC. It is optionally possible to restrict access to the Redshift cluster to MicroStrategy by creating a Customer VPC. A Security Group further restricts access to Redshift
• #28: Amazon CloudWatch monitors your Amazon Web Services (AWS) resources and the applications you run on AWS in real-time. You can use CloudWatch to collect and track metrics, which are the variables you want to measure for your resources and applications. CloudWatch alarms send notifications or automatically make changes to the resources you are monitoring based on rules that you define.