I have a table with three columns A, B, C, all of type bytea.
There are around 180,000,000 rows in the table. A, B and C all have exactly 20 bytes of data, C sometimes contains NULLs
When creating indexes for all columns with
CREATE INDEX index_A ON transactions USING hash (A);
CREATE INDEX index_B ON transactions USING hash (B);
CREATE INDEX index_C ON transactions USING hash (C);
index_A is created in around 10 minutes, while B and C are taking over 10 hours after which I aborted them. I ran every CREATE INDEX on their own, so no indices were created in parallel. There are also no other queries running in the database.
When running
SELECT * FROM pg_stat_activity;
wait_event_type and wait_event are both NULL, state is active.
Why are the second index creations taking so long, and can I do anything to speed them up?
Ensure the statistics on your table are up-to-date.
Then execute the following query:
SELECT attname, n_distinct, correlation
from pg_stats
where tablename = '<Your table name here>'
Basically, the database will have more work to create indexes when:
The number of distinct values gets higher.
The correlation (= are values in the field physically stored in order) is close to 0.
I suspect you will see field A is different in terms of distinct values and/or a higher correlation than the other 2 fields.
Edit: Basically, creating an index = FULL SCAN of the table and create entries in the index as you progress. With the stats you have shared below that means:
Column A: it was detected as unique
A single scan is enough as the DB knows 1 record = 1 index entry.
Columns B & C : it was detected as having very few distinct values + abs(correlation) is very low.
Each index entry takes an entire FULL SCAN of the table.
Note: the description is simplified to highlight the difference.
Solution 1:
Do not create indexes for B and C.
It might sound stupid but in fact and as explained here, a small correlation means the indexes will probably not be used (an index is useful only when entries are not scattered in all the table blocks).
Solution 2:
Order records on the disk.
The initialization would be something like this:
CREATE TABLE Transactions_order as SELECT * FROM Transactions;
TRUNCATE TABLE Transactions;
INSERT INTO Transactions SELECT * FROM Transactions_order ORDER BY B,C,A;
DROP TABLE Transactions_order;
The tricky part comes next: with insert/update/delete records, you need to keep track of the correlation and ensure it does not drop too much.
If you can't guarantee that, stick to solution 1.
Solution3:
Create partitions and enjoy partition pruning.
There are quite a lot of efforts being made for partitioning recently in postgresql. It could be worth having a look into it.
Related
I have a requirement to transfer data from 2 tables (Table A and Table B) into a new table.
I am using a query to join both A and B tables using an ID column.
Table A and B are archive tables without any indexes. (Millions of records)
Table X and Y are a replica of A and B with good indexes. (Some thousands of records)
Below is the code for my project.
with data as
(
SELECT a.*, b.* FROM A_archive a
join B_archive b where a.transaction_id = b.transaction_id
UNION
SELECT x.*, y.* FROM X x
join Y y where x.transaction_id = y.transaction_id
)
INSERT INTO
Another_Table
(
columns
)
select * from data
On Conflict(transaction_id)
do udpate ...
The above whole thing is running in production environment and has nearly 140 million records.
Due to this production database is taking almost 10 hours to process the data and failing.
I am also having a distributed job scheduler in AWS to schedule this query inside a function and retrieve the latest records every 5 hours. The archive tables store closed invoice data. Pega UI will be using this table for retrieving data about closed invoices and showing to the customer.
Please suggest something that is a bit more performant.
UNION removes duplicate rows. On big unindexed tables that is an expensive operation. Try UNION ALL if you don't need deduplication. It will save the s**tton of data shuffling and comparisons required for deduplication.
Without indexes on your archival tables your JOIN operation will be grossly inefficient. Index, at a minimum, the transaction_id columns you use in your ON clause.
You don't say what you want to do with the resulting table. In many cases you'll be able to use a VIEW rather than a table for your purposes. A VIEW removes the work of creating the derived table. Actually it defers the work to the time of SELECT operations using the derived structure. If your SELECT operations have highly selective WHERE clauses the savings can be astonishing. For this to work well you may need to put appropriate indexes on your archival tables.
You use SELECT * when you could enumerate the columns you need. That certainly puts one redundant column into your result: it generates two copies of transaction_id. It also may generate other redundant or unused data. Always avoid SELECT * in production software unless you know you need it.
Keep this in mind: SQL is declarative, not procedural. You declare (describe) the result you require, and you let the server work out the best way to get it. VIEWs let the server do this work for you in cases like your table combination. It will use the indexes you provide as best it can.
That UNION must be costly, it pretty much builds a temp-table in the background containing all the A-B + X-Y records, sorts it (over all fields) and then removes any doubles. If you say 100 million records are involved then that's a LOT of sorting going on that most likely will involve swapping out to disk.
Keep in mind that you only need to do this if there are expected duplicates
in the result from the JOIN between A and B
in the result from the JOIN between X and Y
in the combined result from the two above
IF neither of those are expected, just use UNION ALL
In fact, in that case, why not have 1 INSERT operation for A-B and another one for X-Y? Going by the description I'd say that whatever is in X-Y should overrule whatever is in A-B anyway, right?
Also, as mentioned by O.Jones, archive tables or not, they should come at least with a (preferably clustered) index on the transaction_id fields you're JOINing on. (same for the Another_Table btw)
All that said, processing 100M records in 1 transaction IS going to take some time, it's just a lot of data that's being moved around. But 10h does sound excessive indeed.
I'm currently working on a benchmark (which is part of my bachelor thesis) that compares SQL and NoSQL Databases based on an abstract data model an abstract queries to achieve fair implementation on all systems.
I'm currently working on the implementation of a query that is specified as follows:
I have a table in Cassandra that is specified as follows:
CREATE TABLE allocated(
partition_key int,
financial_institution varchar,
primary_uuid uuid,
report_name varchar,
view_name varchar,
row_name varchar,
col_name varchar,
amount float,
PRIMARY KEY (partition_key, report_name, primary_uuid));
This table contains about 100,000,000 records (~300GB).
We now need to calculate the sum for the field "amount" for every possible combination of report_name, view_name, col_name and row_name.
In SQL this would be quite easy, just select sum (amount) and group it by the fields you want.
However, since Cassandra does not support these operations (which is perfectly fine) I need to achieve this on another way.
Currently I achieve this by doing a full-table walk, processing each record and storing the sum in a HashMap in Java for each combination.
The prepared statement I use is as follows:
SELECT
partition_key,
financial_institution,
report_name,
view_name,
col_name,
row_name,
amount
FROM allocated;
That works partially on machines with lots on RAM for both, cassandra and the Java app, but crashes on smaller machines.
Now I'm wondering whether it's possible to achieve this on a faster way?
I could imagine using the partition_key, which serves also as the cassandra partition key and do this for every partition (I have 5 of them).
Also I though of doing this multithreaded by assigning every partition and report to a seperate thread and running it parallel. But I guess this would cause a lot of overhead on the application side.
Now to the actual question: Would you recommend another execution strategy to achieve this?
Maybe I still think too much in a SQL-like way.
Thank you for you support.
Here are two ideas that may help you.
1) You can efficiently scan rows in any table using the following approach. Consider a table with PRIMARY KEY (pk, sk, tk). Let's use a fetch size of 1000, but you can try other values.
First query (Q1):
select whatever_columns from allocated limit 1000;
Process these and then record the value of the three columns that form the primary key. Let's say these values are pk_val, sk_val, and tk_val. Here is your next query (Q2):
select whatever_columns from allocated where token(pk) = token(pk_val) and sk = sk_val and tk > tk_val limit 1000;
The above query will look for records for the same pk and sk, but for the next values of tk. Keep repeating as long as you keep getting 1000 records. When get anything less, you ignore the tk, and do greater on sk. Here is the query (Q3):
select whatever_columns from allocated where token(pk) = token(pk_val) and sk > sk_val limit 1000;
Again, keep doing this as long as you get 1000 rows. Once you are done, you run the following query (Q4):
select whatever_columns from allocated where token(pk) > token(pk_val) limit 1000;
Now, you again use the pk_val, sk_val, tk_val from the last record, and run Q2 with these values, then Q3, then Q4.....
You are done when Q4 returns less than 1000.
2) I am assuming that 'report_name, view_name, col_name and row_name' are not unique and that's why you maintain a hashmap to keep track of the total amount whenever you see the same combination again. Here is something that may work better. Create a table in cassandra where key is a combination of these four values (maybe delimited). If there were three, you could have simply used a composite key for those three. Now, you also need a column called amounts which is a list. As you are scanning the allocate table (using the approach above), for each row, you do the following:
update amounts_table set amounts = amounts + whatever_amount where my_primary_key = four_col_values_delimited;
Once you are done, you can scan this table and compute the sum of the list for each row you see and dump it wherever you want. Note that since there is only one key, you can scan using only token(primary_key) > token(last_value_of_primary_key).
Sorry if my description is confusing. Please let me know if this helps.
I have a table in postgresql that contains an array which is updated constantly.
In my application i need to get the number of rows for which a specific parameter is not present in that array column. My query looks like this:
select count(id)
from table
where not (ARRAY['parameter value'] <# table.array_column)
But when increasing the amount of rows and the amount of executions of that query (several times per second, possibly hundreds or thousands) the performance decreses a lot, it seems to me that the counting in postgresql might have a linear order of execution (I’m not completely sure of this).
Basically my question is:
Is there an existing pattern I’m not aware of that applies to this situation? what would be the best approach for this?
Any suggestion you could give me would be really appreciated.
PostgreSQL actually supports GIN indexes on array columns. Unfortunately, it doesn't seem to be usable for NOT ARRAY[...] <# indexed_col, and GIN indexes are unsuitable for frequently-updated tables anyway.
Demo:
CREATE TABLE arrtable (id integer primary key, array_column integer[]);
INSERT INTO arrtable(1, ARRAY[1,2,3,4]);
CREATE INDEX arrtable_arraycolumn_gin_arr_idx
ON arrtable USING GIN(array_column);
-- Use the following *only* for testing whether Pg can use an index
-- Do not use it in production.
SET enable_seqscan = off;
explain (buffers, analyze) select count(id)
from arrtable
where not (ARRAY[1] <# arrtable.array_column);
Unfortunately, this shows that as written we can't use the index. If you don't negate the condition it can be used, so you can search for and count rows that do contain the search element (by removing NOT).
You could use the index to count entries that do contain the target value, then subtract that result from a count of all entries. Since counting all rows in a table is quite slow in PostgreSQL (9.1 and older) and requires a sequential scan this will actually be slower than your current query. It's possible that on 9.2 an index-only scan can be used to count the rows if you have a b-tree index on id, in which case this might actually be OK:
SELECT (
SELECT count(id) FROM arrtable
) - (
SELECT count(id) FROM arrtable
WHERE (ARRAY[1] <# arrtable.array_column)
);
It's guaranteed to perform worse than your original version for Pg 9.1 and below, because in addition to the seqscan your original requires it also needs an GIN index scan. I've now tested this on 9.2 and it does appear to use an index for the count, so it's worth exploring for 9.2. With some less trivial dummy data:
drop index arrtable_arraycolumn_gin_arr_idx ;
truncate table arrtable;
insert into arrtable (id, array_column)
select s, ARRAY[1,2,s,s*2,s*3,s/2,s/4] FROM generate_series(1,1000000) s;
CREATE INDEX arrtable_arraycolumn_gin_arr_idx
ON arrtable USING GIN(array_column);
Note that a GIN index like this will slow updates down a LOT, and is quite slow to create in the first place. It is not suitable for tables that get updated much at all - like your table.
Worse, the query using this index takes up to twice times as long as your original query and at best half as long on the same data set. It's worst for cases where the index is not very selective like ARRAY[1] - 4s vs 2s for the original query. Where the index is highly selective (ie: not many matches, like ARRAY[199]) it runs in about 1.2 seconds vs the original's 3s. This index simply isn't worth having for this query.
The lesson here? Sometimes, the right answer is just to do a sequential scan.
Since that won't do for your hit rates, either maintain a materialized view with a trigger as #debenhur suggests, or try to invert the array to be a list of parameters that the entry does not have so you can use a GiST index as #maniek suggests.
Is there an existing pattern I’m not aware of that applies to this
situation? what would be the best approach for this?
Your best bet in this situation might be to normalize your schema. Split the array out into a table. Add a b-tree index on the table of properties, or order the primary key so it's efficiently searchable by property_id.
CREATE TABLE demo( id integer primary key );
INSERT INTO demo (id) SELECT id FROM arrtable;
CREATE TABLE properties (
demo_id integer not null references demo(id),
property integer not null,
primary key (demo_id, property)
);
CREATE INDEX properties_property_idx ON properties(property);
You can then query the properties:
SELECT count(id)
FROM demo
WHERE NOT EXISTS (
SELECT 1 FROM properties WHERE demo.id = properties.demo_id AND property = 1
)
I expected this to be a lot faster than the original query, but it's actually much the same with the same sample data; it runs in the same 2s to 3s range as your original query. It's the same issue where searching for what is not there is much slower than searching for what is there; if we're looking for rows containing a property we can avoid the seqscan of demo and just scan properties for matching IDs directly.
Again, a seq scan on the array-containing table does the job just as well.
I think with Your current data model You are out of luck. Try to think of an algorithm that the database has to execute for Your query. There is no way it could work without sequential scanning of data.
Can You arrange the column so that it stores the inverse of data (so that the the query would be select count(id) from table where ARRAY[‘parameter value’] <# table.array_column) ? This query would use a gin/gist index.
I have a table briefly structured like this:
tn( id integer NOT NULL primary key DEFAULT nextval('tn_sequence'),
create_dt TIMESTAMP NOT NULL DEFAULT NOW(),
...............
deleted boolean );
create_dt is the timestamp when the row is inserted into the database.
deleted indicates that the row is or no longer useful.
And I have the following queries:
select * from tn where create_dt > ( NOW() - interval '150 seconds ) and deleted = FALSE;
select * from tn where create_dt < ( NOW() - interval '150 seconds ) and deleted = FALSE;
My question is how these query will slow down when the number of rows increase? For instance, when the number of rows exceeds 10K, 20K, or 100K, will it make a big impact on the speed? Is there any way I can optimize these queries? Note that every 5 seconds I will turn the column 'deleted' of rows which are older than 150 seconds into 'TRUE'.
The effect of table growth on performance will depend on the query plan chosen, available indexes, the selectivity of the query, and lots of other factors. EXPLAIN ANALYZE on the query might help. In short, if your query only selects a few rows and can use a simple b-tree index then it won't usually slow down tons, only a little as the index grows. On the other hand queries using complex non-indexed conditions or returning lots of rows could perform very badly indeed.
Your issue appears to mirror that in the question How should we handle rows which won't be queried once they are old in PostgreSQL?
The advice given there should apply:
Use a partial index with the condition WHERE (not deleted); or
partition on 'deleted' with constraint exclusion enabled.
For example, you might:
CREATE INDEX create_dt_when_not_deleted_idx
ON tn (create_dt)
WHERE (NOT deleted);
This includes only rows where deleted = 'f' (assuming deleted is `not null) in the index. This isn't the same as having them gone from the table completely.
Nothing changes with full table sequential scans, the deleted='t' rows must still be scanned; and
There's more I/O than if the deleted = 't' rows weren't there because any given heap page is likely to contain a mix of deleted = 't' and deleted = 'f' rows.
You can reduce the impact of the latter by CLUSTERing on an index that includes deleted. Again, this will have no effect on sequential scans. To help with sequential scans you would have to partition the table on deleted.
Pg 9.2's index only scans should (I think, haven't tested) use the partial index. When an index only scan is possible the partial index should be as fast as an index on a table containing only the deleted = 'f' rows.
Note that you'll need to keep table and index bloat under control. Ensure autovaccum runs very frequently and use a current version of PostgreSQL that doesn't need things like manually-managed free space map and has the latest, best-behaved autovacuum. I'd recommend 9.0 or above, preferably 9.1 or 9.2. Tune autovacuum to run aggressively.
When tuning and testing performance - test your queries with EXPLAIN ANALYZE, don't just guess.
Am I safe to assume that where I have stored procedures using the tempdb to write a temporary table, I'd be better off switching these to table variables to get better performance?
Temp tables are better in performance. If you use a Table Variable and the Data in the Variable gets too big, the SQL Server converts the Variable automatically into a temp table.
It depends, like almost every Database related question, on what you try to do. So it is hard to answer without more information.
So my answer is, try it and have a look at the execution plan. Use the fastest way with the lowest costs.
MSDN - Displaying Graphical Execution Plans (SQL Server Management Studio)
#Table can be faster as there is less "setup time" since the object is in memory only.
#Tables have a lot of catches though.
You can have a primary key on a #Table but thats about it. Other indexes Clustered NonClustered for combinations of columns are not possible.
Also if your table is going to contain any real data volumes (more then about 200 maybe 1000 rows) then accessing the table will be slower. Especially when you will probably not have a useful index on it.
#Tables are a pain in procs as they need to be dropped when debugging, They take longer to create. and they take longer to setup as you need to add indexs as a second step. But if you have lots of data then its #tables every time.
Even in cases where you have less then 100 rows of data in a table you may still want to use #Tables as you can create a usefull index on the table.
In summary i use #Tables most of the time for the ease when doing simple proc etc. But anything that need to perform should be a #Table.
#Tables have no statistics so the execution plan entails more guesswork. Hence the recommended upper limit of 1000-ish rows. #Tables have statistics but these can be cached between invocations. If your cardinalities differ significantly each time the SP runs you'd want to REBUILD and RECOMPILE each time. This is an overhead, of course, but one which must be balanced against the cost of a rubbish plan.
Both types will do IO to TempDB.
So no, #Tables are not a panacea.
Table variables can perform very poorly as the number of rows in them increases.
Why is this?
Table variables don’t have distribution statistics and don’t trigger recompiles. Because of this, SQL Server is not able to estimate the number of rows in a table variable like it does for normal tables. When the optimiser compiles code that contains a table variable, it assumes a table is empty and uses an expected row count of 1 for the cardinality estimate. Because the optimiser only thinks a table variable contains a single row, it picks operators for the execution plan that work well with a small set of records, like the NESTED LOOPS operator for a JOIN operation.
As an example, I have just fixed a stored procedure which was performing poorly. The code was populating a table variable and using it in a join to filter the number of rows to accounts which were relevant:
FROM dbo.DimInvestorAccount
INNER JOIN #accounts acclist
ON acclist.AccountNumber = DimInvestorAccount.investorAccountNumber
+ 9 additional tables joined...
When run for list of 1700 accounts, the query was taking 1m17s. Just changing the filter table definition from:
DECLARE #accounts TABLE (AccountNumber VARCHAR(20) COLLATE Latin1_General_BIN INDEX idx NONCLUSTERED)
to
CREATE TABLE #accounts (AccountNumber VARCHAR(20) COLLATE Latin1_General_BIN INDEX idx NONCLUSTERED)
brought the query time down to 800ms. Note that with 5 rows in the table, there was no significant difference - both temp table and table variable run in +/-400ms.
Microsoft's recommendation is to use Table Variables if the number of rows is <100.
Note that Microsoft have made changes in SQL Server 2019 to improve this (v15.x/Compatibility level 150)