Cluster key selection on Snowflake - snowflake-cloud-data-platform

(Submitting on behalf of a Snowflake User)
QUESTION:
Why would the filter or the search key(key used in where clause) would be a better choice for cluster key than an order by or group by key.
One resource recommends reading: https://support.snowflake.net/s/article/case-study-how-clustering-can-improve-your-query-performance
Another resource mentions:
The performance of query filter will be better because the data is sorted it would skip all the rows which are not required.
For the scenario which has query filter on columns which are not part of sort order but the columns in group by and order by are part of data sort order (clustered keys), it may take time to select those data but the sorting would be easy since the data is already in an order.
A 3rd resource states:
The clustering key is important for the WHERE clause when you only select a small portion of the overall data that you have in your tables, because it can reduce the amount of data that has to be read from the Storage into the Compute when the Optimizer can use the clustering key for Query Pruning.
You can alternatively use the clustering key to optimize table inserts and possibly also query output (eg sort order).
Your choice should depend on your priorities, there is no cure all unless a single key covers all above.
To which the User responds with the following questions:
If I always insert the rows in the order in which they will be retrieved, do I still need to create a cluster key? For example if a table is always queried using a date_timestamp and if I ensure that I am inserting in the table order by date_timestamp, do I still need to create a cluster key on date_timestamp?
Any thoughts, recommendations, etc.? Thanks!

For choosing a cluster key based on FILTER/GROUP/SORT. The first "resource" is right.
If the filter will result in pruning, then it is probably best (so that data can be skipped.) If all/most of the data must be read, then clustering on a GROUP/SORT key is probably fast (so less time is spent re-sorting) These docs state:
Typically, queries benefit from clustering when the queries filter or
sort on the clustering key for the table. Sorting is commonly done for
ORDER BY operations, for GROUP BY operations, and for some joins.
For the second question on natural clustering, there would be little to no performance benefit for defining a cluster key in that case.

Related

How should I setup my DynamoDB keys for an activity stream table

I'm wondering what the best way to setup the keys for a table holding activity stream data. Each activity type will have different attributes (with some common ones). Here is an example of what some items will consist of:
A follow activity:
type
user_id
timestamp
follower_user_id
followee_user_id
A comment activity
type
user_id
timestamp
comment_id
commenter_user_id
commented_user_id
For displaying the stream I will be querying against the user_id and ordering by timestamp. There will also be other types of queries - for example I will occasionally need to query user_id AND type as well as stuff like comment_id, follower_user_id etc.
So my questions are:
Should my primary key be a hash and range key using user_id and timestamp?
Do I need secondary indexed for every other item - e.g. comment_id or will results return quick enough without the index? Secondary indexes are limited to 5 which wouldn't be enough for all the types of queries I will need to perform.
I'd consider whether you could segment the data into two (or more) tables - allowing better use of your queries. Combine the two as (and if) needed, ie - your type becomes your table rather than a discriminator like you would do in SQL
If you don't separate the tables, then my answers would be
Yes - I think that would be the best bet given that it seems like most of the time, that will be the way you are using it.
No. But you do need to consider what the most frequent queries are and the performance considerations around it. Which ones need to be performant - and which ones are "good enough" good enough?
A combination of caching and asynchronous processing can allow a slow performing scan to be good enough - but it doesn't eliminate the requirement to have some local secondary indexes.

Clustered index consideration in regards to distinct valus and large result sets and a single vertical table for auditing

I've been researching best practices for creating clustered indexes and I'm just trying to totally understand these two suggestions that's listed with pretty much every BLOG or article on the matter
Columns that contain a large number of distinct values.
Queries that return large result sets.
These seem to be slightly contrary or I'm guessing maybe it just depends on how you're accessing the table.. Or my interpretation of what "large result sets" mean is wrong....
Unless you're doing range queries over the clustered column it seems like you typically won't be getting large result sets that matter. So in cases where SQL Server defaults the clustered indexes on the PK you're rarely going to fulfill the large result set suggestion but of course it does the large number of distinct values..
To give the question a little more context. This quetion stems from a vertical auditing table we have that has a column for TABLE.... Every single query that's written against this table has a
WHERE TABLE = 'TABLENAME'
But the TableName is highly non distinct... Each result set of tablenames is rather large which seems to fulfill that second conditon but it's definitely not largerly unique.... Which means all that other stuff happens with having to add the 4 byte Uniquifer (sp?) which makes the table a lot larger etc...
This situation has come up a few times for me when I've come upon DBs that have say all the contact or some accounts normalized into a single table and they are only separated by a TYPE parameter. Which is on every query....
In the case of the audit table the queries are typically not that exciting either they are just sorted by date modified, sometimes filtered by column, user that made the change etc...
My other thought with this auditing scenario was to just make the auditing table a HEAP so that inserting is fast so there's not contention between tables being audited and then to generate indexed views over the data ...
Index design is just as much art as it is science.
There are many things to consider, including:
How the table will be accessed most often: mostly inserts? any updates? more SELECTs than DML statements? Any audit table will likely have mostly inserts, no updates, rarely deletes unless there is a time-limit on the data, and some SELECTs.
For Clustered indexes, keep in mind that the data in each column of the clustered index will be copied into each non-clustered index (though not for UNIQUE indexes, I believe). This is helpful as those values are available to queries using the non-clustered index for covering, etc. But it also means that the physical space taken up by the non-clustered indexes will be that much larger.
Clustered indexes generally should either be declared with the UNIQUE keyword or be the Primary Key (though there are exceptions, of course). A non-unique clustered index will have a hidden 4-byte field called a uniqueifier that is required to make each row with a non-unique key value addressable, and is just wasted space given that the order of your rows within the non-unique groupings is not apparently obvious so trying to narrow down to a single row is still a range.
As is mentioned everywhere, the clustered index is the physical ordering of the data so you want to cater to what needs the best I/O. This relates also to the point directly above where non-unique clustered indexes have an order but if the data is truly non-unique (as opposed to unique data but missing the UNIQUE keyword when the index was created) then you miss out on a lot of the benefit of having the data physically ordered.
Regardless of any information or theory, TEST TEST TEST. There are many more factors involved that pertain to your specific situation.
So, you mentioned having a Date field as well as the TableName. If the combination of the Date and TableName is unique then those should be used as a composite key on a PK or UNIQUE CLUSTERED index. If they are not then find another field that creates the uniqueness, such as UserIDModified.
While most recommendations are to have the most unique field as the first one (due to statistics being only on the first field), this doesn't hold true for all situations. Given that all of your queries are by TableName, I would opt for putting that field first to make use of the physical ordering of the data. This way SQL Server can read more relevant data per read without having to seek to other locations on disk. You would likely also being ordering on the Date so I would put that field second. Putting TableName first will cause higher fragmentation across INSERTs than putting the Date first, but upon an index rebuild the data access will be faster as the data is already both grouped ( TableName ) and ordered ( Date ) as the queries expect. If you put Date first then the data is still ordered properly but the rows needed to satisfy the query are likely spread out across the datafile(s) which would require more I/O to get. AND, more data pages to satisfy the same query means more pages in the Buffer Pool, potentially pushing out other pages and reducing Page Life Expectancy (PLE). Also, you would then really need to inculde the Date field in all queries as any queries using only TableName (and possibly other filters but NOT using the Date field) will have to scan the clustered index or force you to create a nonclustered index with TableName being first.
I would be weary of the Heap plus Indexed View model. Yes, it might be optimized for the inserts but the system still needs to maintain the data in the indexed view across all DML statements against the heap. Again you would need to test, but I don't see that being materially better than a good choice of fields for a clustered index on the audit table.

db indexes application - best practice

When developing large systems (hundreds of tables) Do you create the indexes (and to lesser extend the other constraints in the DB) when you create the entities (tables), or wait for the system to be running (may be private Beta) to decide where to put the indexes?
I design indexes based on the eventual query scenarios. What will be the most common queries run against the table? That should inform index design - both to optimize query performance as well as to minimize insert/update/delete overhead.
Simply creating a clustered index on the primary key, for example, may make sense in a theoretical world up front, but may not mirror real-world query load.
For example: what if you have a table of order items, where 0-n order items are associated with a parent order? Do you just create an order item ID column, designate it the primary key, and burn your clustered index even though in the real world, 90% of your query activity against this table will be "get order items for order xyz", implying that a clustered index on parent order ID might make more sense than the "default" primary key clustered index on order item ID?
You can do a lot of this up front by knowing what scenarios your application will enable. Then, you can also do traces in the real world and analyze them to find where you are missing indexes; SQL Server, for example, ships with tools to do this, there are third-party tools too. One technique I use sometimes is also to do a big trace, upload the trace info into a table, and query it for distinct SQL statements (based on whatever criteria... e.g. give me all UPDATEs against table xyz...) and then you can do a query plan for those statements and see how good your indexing is by, for example, looking for and addressing table or index scans appropriately - and verifying by re-examining the execution plan for the query.
Some cautionary notes... don't apply indexes willy-nilly based on traces. An index on a table will affect overall performance of all queries against the table. Don't assume that a table or index scan (rather than a seek) is necessarily bad; it doesn't matter in a ten-row table. Index optimization is a combination of science and art, so keeping it simple is critical, testing frequently after small incremental changes is a good way to retain sanity and be able to roll back frequently, and above all, when you have a set of changes, script them out so that your DBA has an exact protocol of what will be done, and can easily determine where/what to roll back if needed.
If you know what fields you are going to be using most of the time (where and order by clauses`), you might as well create them when creating the entities.
You can always revisit later, and any DBA worth his salt would.

Approaches to table partitioning in SQL Server

The database I'm working with is currently over 100 GiB and promises to grow much larger over the next year or so. I'm trying to design a partitioning scheme that will work with my dataset but thus far have failed miserably. My problem is that queries against this database will typically test the values of multiple columns in this one large table, ending up in result sets that overlap in an unpredictable fashion.
Everyone (the DBAs I'm working with) warns against having tables over a certain size and I've researched and evaluated the solutions I've come across but they all seem to rely on a data characteristic that allows for logical table partitioning. Unfortunately, I do not see a way to achieve that given the structure of my tables.
Here's the structure of our two main tables to put this into perspective.
Table: Case
Columns:
Year
Type
Status
UniqueIdentifier
PrimaryKey
etc.
Table: Case_Participant
Columns:
Case.PrimaryKey
LastName
FirstName
SSN
DLN
OtherUniqueIdentifiers
Note that any of the columns above can be used as query parameters.
Rather than guess, measure. Collect statistics of usage (queries run), look at the engine own statistics like sys.dm_db_index_usage_stats and then you make an informed decision: the partition that bests balances data size and gives best affinity for the most often run queries will be a good candidate. Of course you'll have to compromise.
Also don't forget that partitioning is per index (where 'table' = one of the indexes), not per table, so the question is not what to partition on, but which indexes to partition or not and what partitioning function to use. Your clustered indexes on the two tables are going to be the most likely candidates obviously (not much sense to partition just a non-clustered index and not partition the clustered one) so, unless you're considering redesign of your clustered keys, the question is really what partitioning function to choose for your clustered indexes.
If I'd venture a guess I'd say that for any data that accumulates over time (like 'cases' with a 'year') the most natural partition is the sliding window.
If you have no other choice you can partition by key module the number of partition tables.
Lets say that you want to partition to 10 tables.
You will define tables:
Case00
Case01
...
Case09
And partition you data by UniqueIdentifier or PrimaryKey module 10 and place each record in the corresponding table (Depending on your unique UniqueIdentifier you might need to start manual allocation of ids).
When performing a query, you will need to run same query on all tables, and use UNION to merge the result set into a single query result.
It's not as good as partitioning the tables based on some logical separation which corresponds to the expected query, but it's better then hitting the size limit of a table.
Another possible thing to look at (before partitioning) is your model.
Are you in a normalized database? Are there further steps which could improve performance by different choices in the normalization/de-/partial-normalization? Are there options to transform the data into a Kimball-style dimensional star model which is optimal for reporting/querying?
If you aren't going to drop partitions of the table (sliding window, as mentioned) or treat different partitions differently (you say any columns can be used in the query), I'm not sure what you are trying to get out of the partitioning that you won't already get out of your indexing strategy.
I'm not aware of any table limits on rows. AFAIK, the number of rows is limited only by available storage.

What columns generally make good indexes?

As a follow up to "What are indexes and how can I use them to optimise queries in my database?" where I am attempting to learn about indexes, what columns are good index candidates? Specifically for an MS SQL database?
After some googling, everything I have read suggests that columns that are generally increasing and unique make a good index (things like MySQL's auto_increment), I understand this, but I am using MS SQL and I am using GUIDs for primary keys, so it seems that indexes would not benefit GUID columns...
Indexes can play an important role in query optimization and searching the results speedily from tables. The most important step is to select which columns are to be indexed. There are two major places where we can consider indexing: columns referenced in the WHERE clause and columns used in JOIN clauses. In short, such columns should be indexed against which you are required to search particular records. Suppose, we have a table named buyers where the SELECT query uses indexes like below:
SELECT
buyer_id /* no need to index */
FROM buyers
WHERE first_name='Tariq' /* consider indexing */
AND last_name='Iqbal' /* consider indexing */
Since "buyer_id" is referenced in the SELECT portion, MySQL will not use it to limit the chosen rows. Hence, there is no great need to index it. The below is another example little different from the above one:
SELECT
buyers.buyer_id, /* no need to index */
country.name /* no need to index */
FROM buyers LEFT JOIN country
ON buyers.country_id=country.country_id /* consider indexing */
WHERE
first_name='Tariq' /* consider indexing */
AND
last_name='Iqbal' /* consider indexing */
According to the above queries first_name, last_name columns can be indexed as they are located in the WHERE clause. Also an additional field, country_id from country table, can be considered for indexing because it is in a JOIN clause. So indexing can be considered on every field in the WHERE clause or a JOIN clause.
The following list also offers a few tips that you should always keep in mind when intend to create indexes into your tables:
Only index those columns that are required in WHERE and ORDER BY clauses. Indexing columns in abundance will result in some disadvantages.
Try to take benefit of "index prefix" or "multi-columns index" feature of MySQL. If you create an index such as INDEX(first_name, last_name), don’t create INDEX(first_name). However, "index prefix" or "multi-columns index" is not recommended in all search cases.
Use the NOT NULL attribute for those columns in which you consider the indexing, so that NULL values will never be stored.
Use the --log-long-format option to log queries that aren’t using indexes. In this way, you can examine this log file and adjust your queries accordingly.
The EXPLAIN statement helps you to reveal that how MySQL will execute a query. It shows how and in what order tables are joined. This can be much useful for determining how to write optimized queries, and whether the columns are needed to be indexed.
Update (23 Feb'15):
Any index (good/bad) increases insert and update time.
Depending on your indexes (number of indexes and type), result is searched. If your search time is gonna increase because of index then that's bad index.
Likely in any book, "Index Page" could have chapter start page, topic page number starts, also sub topic page starts. Some clarification in Index page helps but more detailed index might confuse you or scare you. Indexes are also having memory.
Index selection should be wise. Keep in mind not all columns would require index.
Some folks answered a similar question here: How do you know what a good index is?
Basically, it really depends on how you will be querying your data. You want an index that quickly identifies a small subset of your dataset that is relevant to a query. If you never query by datestamp, you don't need an index on it, even if it's mostly unique. If all you do is get events that happened in a certain date range, you definitely want one. In most cases, an index on gender is pointless -- but if all you do is get stats about all males, and separately, about all females, it might be worth your while to create one. Figure out what your query patterns will be, and access to which parameter narrows the search space the most, and that's your best index.
Also consider the kind of index you make -- B-trees are good for most things and allow range queries, but hash indexes get you straight to the point (but don't allow ranges). Other types of indexes have other pros and cons.
Good luck!
It all depends on what queries you expect to ask about the tables. If you ask for all rows with a certain value for column X, you will have to do a full table scan if an index can't be used.
Indexes will be useful if:
The column or columns have a high degree of uniqueness
You frequently need to look for a certain value or range of values for
the column.
They will not be useful if:
You are selecting a large % (>10-20%) of the rows in the table
The additional space usage is an issue
You want to maximize insert performance. Every index on a table reduces insert and update performance because they must be updated each time the data changes.
Primary key columns are typically great for indexing because they are unique and are often used to lookup rows.
Any column that is going to be regularly used to extract data from the table should be indexed.
This includes:
foreign keys -
select * from tblOrder where status_id=:v_outstanding
descriptive fields -
select * from tblCust where Surname like "O'Brian%"
The columns do not need to be unique. In fact you can get really good performance from a binary index when searching for exceptions.
select * from tblOrder where paidYN='N'
In general (I don't use mssql so can't comment specifically), primary keys make good indexes. They are unique and must have a value specified. (Also, primary keys make such good indexes that they normally have an index created automatically.)
An index is effectively a copy of the column which has been sorted to allow binary search (which is much faster than linear search). Database systems may use various tricks to speed up search even more, particularly if the data is more complex than a simple number.
My suggestion would be to not use any indexes initially and profile your queries. If a particular query (such as searching for people by surname, for example) is run very often, try creating an index over the relevate attributes and profile again. If there is a noticeable speed-up on queries and a negligible slow-down on insertions and updates, keep the index.
(Apologies if I'm repeating stuff mentioned in your other question, I hadn't come across it previously.)
It really depends on your queries. For example, if you almost only write to a table then it is best not to have any indexes, they just slow down the writes and never get used. Any column you are using to join with another table is a good candidate for an index.
Also, read about the Missing Indexes feature. It monitors the actual queries being used against your database and can tell you what indexes would have improved the performace.
Your primary key should always be an index. (I'd be surprised if it weren't automatically indexed by MS SQL, in fact.) You should also index columns you SELECT or ORDER by frequently; their purpose is both quick lookup of a single value and faster sorting.
The only real danger in indexing too many columns is slowing down changes to rows in large tables, as the indexes all need updating too. If you're really not sure what to index, just time your slowest queries, look at what columns are being used most often, and index them. Then see how much faster they are.
Numeric data types which are ordered in ascending or descending order are good indexes for multiple reasons. First, numbers are generally faster to evaluate than strings (varchar, char, nvarchar, etc). Second, if your values aren't ordered, rows and/or pages may need to be shuffled about to update your index. That's additional overhead.
If you're using SQL Server 2005 and set on using uniqueidentifiers (guids), and do NOT need them to be of a random nature, check out the sequential uniqueidentifier type.
Lastly, if you're talking about clustered indexes, you're talking about the sort of the physical data. If you have a string as your clustered index, that could get ugly.
A GUID column is not the best candidate for indexing. Indexes are best suited to columns with a data type that can be given some meaningful order, ie sorted (integer, date etc).
It does not matter if the data in a column is generally increasing. If you create an index on the column, the index will create it's own data structure that will simply reference the actual items in your table without concern for stored order (a non-clustered index). Then for example a binary search can be performed over your index data structure to provide fast retrieval.
It is also possible to create a "clustered index" that will physically reorder your data. However you can only have one of these per table, whereas you can have multiple non-clustered indexes.
The ol' rule of thumb was columns that are used a lot in WHERE, ORDER BY, and GROUP BY clauses, or any that seemed to be used in joins frequently. Keep in mind I'm referring to indexes, NOT Primary Key
Not to give a 'vanilla-ish' answer, but it truly depends on how you are accessing the data
It should be even faster if you are using a GUID.
Suppose you have the records
100
200
3000
....
If you have an index(binary search, you can find the physical location of the record you are looking for in O( lg n) time, instead of searching sequentially O(n) time. This is because you dont know what records you have in you table.
Best index depends on the contents of the table and what you are trying to accomplish.
Taken an example A member database with a Primary Key of the Members Social Security Numnber. We choose the S.S. because the application priamry referes to the individual in this way but you also want to create a search function that will utilize the members first and last name. I would then suggest creating a index over those two fields.
You should first find out what data you will be querying and then make the determination of which data you need indexed.

Resources