Native (PostgreSQL only) streaming data tables

If you want to see (and analyze) only a window of data over some continuous data stream in PostgreSQL, one way is to use a specialized tool like the PipelineDB extension. But if you can't do that, e.g. because you are stuck with AWS RDS or for some other reason, streaming data tables, or continuous views, can be implemented with pretty much PostgreSQL alone.

The basic idea is to have a table that allows for fast INSERT operations, is aggressively VACUUMed, and has some key that can be used to prune outdated entries. This table is fed with the events from the data stream and regularly pruned. Voilà: a streaming data table.

We have done some testing with two approaches on an UNLOGGED table, prune on every INSERT, and pruning at reqular intervals. UNLOGGED is not a problem here, since a view on a data stream can be considered pretty much as ephemeral.

The timed variant is about 5x - 8x faster on INSERTs. And if you balance the timing and the pruning interval right, the window size is almost as stable.

The examples are implemented in Python3 with psycopg2. Putting an index on the table can help or hurt performance, INSERT might get slower but pruning with DELETE faster, depending on the size and structure of the data. Feel free to experiment. In our case, a vanilla BRIN index did just fine.

Instead of using an external scheduler for pruning, like the Python daemon thread in the stream_timed_cleanup.py example, other scheduling mechanisms can be of course used, e.g. pg_cron, or a scheduled Lambda on AWS, or similar.

Feel free to experiment and improve...

MQTT as transport for PostgreSQL events

MQTT has become a de-facto standard for the transport of messages between IoT devices. As a result, a plethora of libraries and MQTT message brokers have become available. Can we use this to transport messages originating from PostgreSQL?

Aa message broker we use Eclipse Mosquitto which is dead simple to set up if you don't have to change the default settings. Such a default installation is neither secure nor highly available, but for our demo it will do just fine. The event generators are written in Python3 with Eclipse paho mqtt for Python.

There are at least two ways to generate events from a PostgreSQL database, pg_recvlogical and NOTIFY / LISTEN. Both have their advantages and shortcomings.

pg_recvlogical:

• Configured on server and database level
• Generates comprehensive information about everything that happens in the database
• No additional programming neccessary
• Needs plugins to decode messages, e.g. into JSON
• Filtering has to be done later, e.g. by the decoder plugin

NOTIFY / LISTEN:

• Configured on DDL and table level
• Generates exactly the information and format you program into the triggers
• Filtering can be done before sending the message
• Needs trigger programming
• The message size is limited to 8000 bytes

Examples for both approaches can be found here. The NOTIFY / LISTEN example lacks a proper decoder but this makes be a good excercise to start with. The pg_recvlogical example needs the wal2json plugin, which can be found here and the proper setup, which is also explained in the Readme. Please note, that the slot used in the example is mqtt_slot, not test_slot:

pg_recvlogical -d postgres --slot mqtt_slot --create-slot -P wal2json

Otherwise, setup.sql should generate all objects to run both examples.

It looks like pgchem::tigress just got a major upgrade

With the Release of PostgreSQL 12.x and OpenBabel 3.x, I decided to see if pgchem::tigress would still compile. Well, it took some minor changes, but YES, it does!

And - it seems like OpenBabel now handles E/Z and enantiomer stereochemistry correctly, at least in SMILES notation. This is a major step forward, but I have to do some more checks before the next release...

Authenticate PostgreSQL users against the Amazon AWS Cognito service

I was asked recently if PostgreSQL could authenticate login users against AWS Cognito.  Since PostgreSQL allows PAM authentication, I was pretty sure it could.

But an (admittedly not exhaustive) search on the web did not produce any PAMs for Cognito.

So I wrote one, using pam-python, boto3, warrant and pyJWT:

cognito_PAM.py

It is designed primarily for PostgreSQL and pgbouncer, so it only supports pam_sm_authenticate and pam_sm_acct_mgmt, and all the work is done in pam_sm_authenticate. Because calling Cognito is comparatively slow, I didn't want to call it twice.

The necessary pam.d config is:

#%PAM-1.0
# Information for PostgreSQL process with the 'pam' option.
auth required  pam_python.so cognito_PAM.py aws_region user_pool_id client_id 
account required pam_python.so cognito_PAM.py

If you use PAM authentication, passwords are sent in cleartext, so transport layer encryption, e.g. SSL/TLS, between client and server becomes mandatory!

I think it does the correct dance of authentication with Cognito and supports USER_SRP_AUTH, but if you see any problem, please raise a paw.

Excel and ODF support for cloudfs_fdw

cloudfs_fdw now supports .xls (Excel 97-2003), .xlsx, and .ods (Open Document Format) Spreadsheets via pandas, xlrd, and odfpy. It requires pandas >= 1.0.1, so Multicorn must be compiled against Python 3.

Since pandas provides sorting and filtering capabilities, cloudfs_fdw tries to push down SQL qualifiers and sort keys when they can be translated into pandas notation.

Take a look and have fun.

cloudfs_fdw

Since I needed a Foreign Data Wrapper for files stored on S3, and the ones I found did things like loading the whole file in memory before sending the first rows, I wrote my own, using Multicorn.

Along the way, I discovered libraries like smart-open and ijson that allow to stream various file formats from various filesystems - and so this escalated a bit, into cloudfs_fdw.

It currently supports CSV and JSON files from S3, HTTP/HTTPS sources and local or network  filesystems but since smart-open supports more than that (e.g. HDFS, SSH), it certainly can be extended if needed.

For now, have fun.

Not all CASTs are created equal?

Can somebody explain this?

[Solved: See the comments section]

PostgreSQL 11.2.

The documentation says:

"A type cast specifies a conversion from one data type to another. PostgreSQL accepts two equivalent syntaxes for type casts:

CAST ( expression AS type )

expression::type

The CAST syntax conforms to SQL; the syntax with :: is historical PostgreSQL usage."

But when I test the lower limits of PostgreSQL's integer types, strange things happen.

select cast(-9223372036854775808 as bigint);
select cast(-2147483648 as integer);
select cast(-32768 as smallint);

All OK.

select -9223372036854775808::bigint;
select -2147483648::integer;
select -32768::smallint;

All fail with SQL Error [22003]: ERROR: out of range

But:

select -9223372036854775807::bigint;
select -2147483647::integer;
select -32767::smallint;

All OK.

???

The Hare and the Hedgehog. Muscle, brain - or both?

In the famous fairy tale the hedgehog wins the race against the hare because he uses his brain to outwit the much faster hare: Brain beats muscle. But is that always the case? And what if we combine the two virtues?

The case at hand: Screening large sets of molecules for chemical simliarity.

Since (sub)graph isomorphism searching faces some mathematical challenges because of nonpolynomial O - even if you can use a specialized index, like pgchem::tigress does - fast similarity searching based on binary fingerprints has gained popularity in recent years.

I was tasked with evaluating a solution to the problem of similarity screening large sets of molecules with PostgreSQL where the fingerprints are generated externally, e.g. with the CDK.

This is, what I came up with...

Preparing the Racetrack

CREATE TABLE cdk.externalfp (
id int4 NOT NULL,
smiles varchar NOT NULL,
pubchemfp varbit NULL,
"cardinality" int4 NULL,
CONSTRAINT externalfp_pk PRIMARY KEY (id)
);

Above is the table definition of the final table. The cardinality column will be not used now, but since it is calculated by the fingerprint generator anyway, keeping it will save some work later. If you want to copy my example code 1:1, please use a database named chemistry and a schema named cdk.

First we need to load some data into the table. I used the free NCISMA99 dataset  from the National Cancer Institute, containing 249081 chemical structures in SMILES notation.

COPY cdk.externalfp (id, smiles) FROM '/tmp/NCISMA99' 
WITH (DELIMITER ' ', HEADER false, FORMAT csv);

And a few seconds later you should have 249081 rows in the table. Now we need to generate the fingerprints. The generator code is here, additionally you need the CDK 2.2 and a PostgreSQL JDBC driver. After changing the code to reflect your JDBC URL you are good to go.

Running the FingerprintGenerator should show no errors and takes about 30 Minutes on my Core i5 Linux Notebook. The fingerprint used is the PubChem fingerprint as described here.
Now we can put an index on the cardinality column (also used later) and are all set.

CREATE INDEX externalfp_cardinality_idx ON cdk.externalfp USING btree (cardinality);

Almost...

The function to calculate the similarity measure is still missing. We use the Tanimoto coefficient, as it is widely used and fairly easy to understand. The Tanimoto coefficient over PostgreSQL BIT VARYING can thus be written in pure SQL as:

CREATE OR REPLACE FUNCTION cdk.tanimoto(bit varying, bit varying)
 RETURNS real
 LANGUAGE sql
 IMMUTABLE STRICT SECURITY INVOKER LEAKPROOF
AS $function$
select length(replace(($1 & $2)::text, '0', ''))::real / length(replace(($1 | $2)::text, '0', ''))::real;
$function$;

Please note that the required bitcount function on BIT VARYING is emulated by removing all 0s and measuring the length of the remaining string, containing all 1s.

The Baseline. No brain, no muscles

For the first naive test, we run FindBySimilarity with the following inputs:

SMILES: CC(=O)OC1=CC=CC=C1C(=O)O
Threshold: 0.9
top N: 10

which gives the following plan from EXPLAIN ANALYZE:

Limit  (cost=14515.22..14516.38 rows=10 width=77) (actual time=6436.022..6436.621 rows=10 loops=1)                                         
  ->  Gather Merge  (cost=14515.22..22587.95 rows=69190 width=77) (actual time=6436.021..6436.618 rows=10 loops=1)                         
        Workers Planned: 2                                                                                                                 
        Workers Launched: 2                                                                                                                
        ->  Sort  (cost=13515.19..13601.68 rows=34595 width=77) (actual time=6432.534..6432.535 rows=8 loops=3)                            
              Sort Key: (((length(replace((('1100000001110000001110000000000000000000000000000000000000000000000000000000000000000000000000
              Sort Method: top-N heapsort  Memory: 27kB                                                                                    
              Worker 0:  Sort Method: quicksort  Memory: 27kB                                                                              
              Worker 1:  Sort Method: top-N heapsort  Memory: 27kB                                                                         
              ->  Parallel Seq Scan on externalfp  (cost=0.00..12767.61 rows=34595 width=77) (actual time=41.329..6432.411 rows=30 loops=3)
                    Filter: (((length(replace((('110000000111000000111000000000000000000000000000000000000000000000000000000000000000000000
                    Rows Removed by Filter: 82997                                                                                          
Planning Time: 0.169 ms                                                                                                                    
Execution Time: 6436.648 ms                                                                                                                                                                                                                                                                                                                           
Including retrieval, this plan leads to an overall response time of about 14 seconds.

First race. Introducing the Hedgehog

An index could be helpful, but can we build one without major programming effort? Yes, due to the work of S. Joshua Swamidass and Pierre Baldi. In their 2007 paper "Bounds and Algorithms for Fast Exact Searches of Chemical Fingerprints in Linear and Sub-Linear Time", they found ways to calculate upper and lower bounds on the cardinality of fingerprints necessary to meet a given similarity once the cardinality of the search fingerprint is known.

The paper covers these calculations for various similarity measures. For Tanimoto it is:

min_cardinality_of_target = floor(cardinality_of_search_argument * similarity_threshold)

and 

max_cardinality_of_target = ceil(cardinality_of_search_argument / similarity_threshold)

The function swamidassBaldiLimitsForTanimoto() in FindBySimilarity calculates those bounds and if you change:

//int[] lohi = swamidassBaldiLimitsForTanimoto(fp.cardinality(), threshold);

int[] lohi = {0, Integer.MAX_VALUE};

to

int[] lohi = swamidassBaldiLimitsForTanimoto(fp.cardinality(), threshold);

//int[] lohi = {0, Integer.MAX_VALUE};

they will be used. Now the index we already put on the "cardinality" column makes sense, allowing the database to filter out impossible candidates before invoking the tanimoto function.

For CC(=O)OC1=CC=CC=C1C(=O)O, the fingerprint has a cardinality of 115, which gives an upper bound of 128 and a lower bound of 103 bits. All fingerprints with cardinalities outside that bounds can be safely ignored since they cannot yield a Tanimoto coefficient >= 0.9.

Now the plan becomes:

Limit  (cost=10445.68..10446.85 rows=10 width=77) (actual time=1492.430..1495.552 rows=10 loops=1)                                                           
  ->  Gather Merge  (cost=10445.68..12328.11 rows=16134 width=77) (actual time=1492.429..1495.548 rows=10 loops=1)                                           
        Workers Planned: 2                                                                                       
        Workers Launched: 2                                                                                                                                  
        ->  Sort  (cost=9445.66..9465.82 rows=8067 width=77) (actual time=1489.566..1489.567 rows=7 loops=3)                                                 
              Sort Key: (((length(replace((('1100000001110000001110000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
              Sort Method: top-N heapsort  Memory: 27kB                                                                                                      
              Worker 0:  Sort Method: top-N heapsort  Memory: 27kB                                                                                           
              Worker 1:  Sort Method: top-N heapsort  Memory: 27kB                                                                                           
              ->  Parallel Bitmap Heap Scan on externalfp  (cost=826.09..9271.33 rows=8067 width=77) (actual time=22.032..1489.502 rows=30 loops=3)          
                    Recheck Cond: ((cardinality >= 103) AND (cardinality <= 128))                                                                            
                    Filter: (((length(replace((('110000000111000000111000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
                    Rows Removed by Filter: 19303                                                                                                            
                    Heap Blocks: exact=1928                                                                                                                  
                    ->  Bitmap Index Scan on externalfp_cardinality_idx  (cost=0.00..821.25 rows=58083 width=0) (actual time=5.238..5.238 rows=58000 loops=1)
                          Index Cond: ((cardinality >= 103) AND (cardinality <= 128))                                                                        
Planning Time: 0.166 ms                                                                                                                                      
Execution Time: 1495.587 ms                                                                                                                                                             

Including retrieval, this plan gives us now an overall response time of about 3 seconds, or 4.6 times faster.

Second race. Introducing the Hare

Time to pull out the big guns. The tanimoto calculation in SQL is apparently slow, primarily because PostgreSQL is lacking a native bit count function for BIT VARYING, so this must be emulated using binary string replace() and length().

However, we can build one in C. Since counting bits is a simple operation for a microprocessor (Actually, it is an art in itself. See Andrew Dalke's popcount benchmark for many different ways to count bits.), a C function should perform much better.

After you installed tanimoto.c, changed the tanimoto function calls in the SQL in FindBySimilarity to tanimoto_c, and disabled the Swamidass/Baldi indexing, we see the raw power of a native function.

The plan now becomes:

Limit  (cost=10363.85..10365.02 rows=10 width=77) (actual time=45.829..48.101 rows=10 loops=1)                                         
  ->  Gather Merge  (cost=10363.85..18436.58 rows=69190 width=77) (actual time=45.827..48.097 rows=10 loops=1)                         
        Workers Planned: 2                                                                                                             
        Workers Launched: 2                                                                                                            
        ->  Sort  (cost=9363.83..9450.32 rows=34595 width=77) (actual time=43.667..43.668 rows=8 loops=3)                              
              Sort Key: (tanimoto_c('11000000011100000011100000000000000000000000000000000000000000000000000000000000000000000000000000
              Sort Method: top-N heapsort  Memory: 27kB                                                                                
              Worker 0:  Sort Method: top-N heapsort  Memory: 27kB                                                                     
              Worker 1:  Sort Method: top-N heapsort  Memory: 27kB                                                                     
              ->  Parallel Seq Scan on externalfp  (cost=0.00..8616.24 rows=34595 width=77) (actual time=0.095..43.559 rows=30 loops=3)
                    Filter: (tanimoto_c('1100000001110000001110000000000000000000000000000000000000000000000000000000000000000000000000
                    Rows Removed by Filter: 82997                                                                                      
Planning Time: 0.062 ms                                                                                                                
Execution Time: 48.145 ms     


Overall response is 0.128 seconds, or 110 times faster.

Muscle beats brain. At least here.

Third race. Relay

But if we enable the Swamidass/Baldi index again and keep the native Tanimoto function, the real magic happens.

The final plan:

Limit  (cost=9427.54..9427.56 rows=10 width=77) (actual time=35.567..35.570 rows=10 loops=1)                                                           
  ->  Sort  (cost=9427.54..9475.94 rows=19361 width=77) (actual time=35.566..35.566 rows=10 loops=1)                                                   
        Sort Key: (tanimoto_c('110000000111000000111000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
        Sort Method: top-N heapsort  Memory: 26kB                                                                                                      
        ->  Bitmap Heap Scan on externalfp  (cost=826.09..9009.15 rows=19361 width=77) (actual time=5.404..35.459 rows=90 loops=1)                     
              Recheck Cond: ((cardinality >= 103) AND (cardinality <= 128))                                                                            
              Filter: (tanimoto_c('11000000011100000011100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
              Rows Removed by Filter: 57910                                                                                                            
              Heap Blocks: exact=6801                                                                                                                  
              ->  Bitmap Index Scan on externalfp_cardinality_idx  (cost=0.00..821.25 rows=58083 width=0) (actual time=4.307..4.307 rows=58000 loops=1)
                    Index Cond: ((cardinality >= 103) AND (cardinality <= 128))                                                                        
Planning Time: 0.149 ms                                                                                                                                
Execution Time: 35.641 ms                                                                                                                                                                                                                                          
                                                                                                       
Final overall response now 0.04 seconds, or 350 times faster!

Muscle + brain: unmatched.

Conclusion

Sometimes the Hare wins, sometimes the Hedgehog does. Always attack nontrivial optimization problems from different angles and experiment, experiment, experiment.

While C largely outperforms the index in this case, it's still good to know both, because using native functions is not always possible, e.g. on an AWS RDS instance. Then, 4.6x faster is better than nothing.

PostgreSQL needs a native builtin function for counting bits in BIT VARYING. Since fingerprint screening can be used not only for chemical structures, but for virtually every complex data that can be broken down into binary features, this would be ever so useful.

Blockchain Soccer

A opinion piece - published on medium:

https://medium.com/@ernst.georg.schmid/blockchain-soccer-869a45a8077b

Wrong prediction of the day

"There is this fear people have that eventually actors are going to be replaced by computer characters. I don't think it is valid at all. The only thing that the technology is going to do is provide the actors with new places to go and new ways to go there."

- Steven Lisberger, 1982, BYTE Magazine, Vol. 07, #11, p. 74

Using PostgreSQL to get things done: 2. Tuning for read performance

Using PostgreSQL to get things done: 1. Installation

Currently I'm doing a short series of videos about data analysis PostgreSQL. Since this is for people who usually do not directly deal with databases every day, it starts very basic with the installation on Linux and Windows.

What do you think, is it worth the effort?

Windows, keep your dirty fingers off my files!

I spent the better part of the morning figuring out why a colleague could not import a PostgreSQL dump in plain format made on Linux on his Windows machine.

According to documentation, this works like so (OS agnostic):

psql dbname < infile

However, this  gave the following error:

ERROR:  missing data for ...

However, the documentation for psql gives an alternative way to read commands from a file:

The -f switch.

 "

Read commands from the file filename, rather than standard input. This option can be repeated and combined in any order with the -c option. When either -c or -f is specified, psql does not read commands from standard input; instead it terminates after processing all the -c and -f options in sequence. Except for that, this option is largely equivalent to the meta-command \i.
If filename is - (hyphen), then standard input is read until an EOF indication or \q meta-command. This can be used to intersperse interactive input with input from files. Note however that Readline is not used in this case (much as if -n had been specified).
Using this option is subtly different from writing psql < filename. In general, both will do what you expect, but using -f enables some nice features such as error messages with line numbers. There is also a slight chance that using this option will reduce the start-up overhead. On the other hand, the variant using the shell's input redirection is (in theory) guaranteed to yield exactly the same output you would have received had you entered everything by hand.

 "


What this doesn't tell you, is that on Windows, CMD.exe apparently somehow tries to interpret the file it reads. And by doing so, it destroyed data in the dump so that COPY was unable to understand it anymore. So the last sentence of the statement above is just theory on Windows.

Long story short, with psql -f all went fine - and don't use I/O redirection with psql on Windows!

Module Monday: safeupdate

A recent tweet pointed me to safeupdate which is a little extension that forbids unconditional UPDATE and DELETE statements, i.e. without WHERE clause.

Let's see how this works...

The build and install procedure is pretty non-spectacular: make, make install and then either activate it per session with load 'safeupdate' or globally by adding it to postgresql.conf:

shared_preload_libraries = 'safeupdate'

After that, if you try an unconditional UPDATE or DELETE on any table, you'll get the following errors:

ERROR:  UPDATE requires a WHERE clause

and

ERROR:  DELETE requires a WHERE clause

respectively and the operation is cancelled.

Unconditional UPDATE and DELETE can still be forced by using a condition that always evaluates to true, like

WHERE TRUE or WHERE 1=1

But now you have to explicitly request this behaviour.

I think this is ever so useful to protect against human error, that I wonder why this isn't something mandatory, even if it's non-standard.

However, there is one glitch: If you JOIN tables to UPDATE or DELETE only certain rows that match the ON condition, this is not recognized as a conditional operation and the WHERE TRUE is still needed to pacify the extension. Well, I can live with that...

A new type for PostgreSQL

CREATE TYPE trumpean AS ENUM ('true', 'alternateTrue');

What happened?

Deutschland	3222
Vereinigte Staaten	3127
Tschechische Republik	2918
Frankreich	66
Vereinigtes Königreich	49
Belgien	34
Polen	23
Slowakei	12
Brasilien	11
Irland	11

Germany suddenly jumped from a two-digit to a four-digit access count?

Sudden FAME! ;-)

One-time passwords with Google Authenticator PAM (and friends)

PostgreSQL allows for more than plain password authentication in pg_hba.conf. One of the most flexible is authenticating against a PAM.

Let's see how this works with one-time passwords from  Google Authenticator.

1.) Install Google Authenticator on your Android or iOS device.

2.) Install the Google Authenticator PAM on the machine where your PostgreSQL server lives, like in Step 1 - 4 of this guide.

3.) Connect your device with the account on that machine.

4.) Create a login role for the database but without a password. Since roles without password can never log in regularly (See the PASSWORD remarks here), this limits this role to PAM logins.

5.) Configure a PAM service for PostgreSQL. E.g. create a file named postgresql where your PAM configs live, on Ubuntu this is /etc/pam.d/. The file should look like this:

auth         sufficient     pam_google_authenticator.so

6.) Configure PostgreSQL to use the PAM. E.g. a line in pg_hba.conf could look like this:

hostssl    all    all    127.0.0.1/32   pam    pamservice=postgresql

And that's basically it. Now, next time you login, PostgreSQL will ask you for a password that is generated individually on your device.

Of course you can use all kinds of PAM with PostgreSQL like this.

Unfortunately, I also found a few caveats along the way. :-(

First, PostgreSQL clients will ask only for one password, regardless if you chain n PAM's for n-factor authentication.

So if you e.g. chain a PAM against LDAP with Google Authenticator as the second factor, this won't work. This seems to be a shortcoming of the PAM implementation in PostgreSQL, not expecting multiple password prompts. It is still possible to enable n-factor authentication though, but only one PAM can prompt for a password. If the other factors are hardware devices like a fingerprint scanner that does not prompt for a password, you are fine.

Alternatively, you can provide your own PAM that takes all passwords in one prompt and handles them internally.

Second, PAM requires PostgreSQL clients to send the password in plaintext. So now is the time to switch on TLS and make it mandatory (Noticed the hostssl switch above?).

Third, some clients like pgAdmin3 break with one-time passwords, because they apparently open new connections without prompting for a password again, but re-use the initial one instead until you disconnect. This obviously does not work with passwords which are valid only for one login attempt.

Stampede is out!

Stampede just got live!

I'll try this for sure, when I have time, when I have time...

pgchem::tigress 4.0 released

pgchem::tigress 4.0 is finally out!

• This builds against PostgreSQL >= 9.4.x and OpenBabel 4.2.x on Linux.
• It contains all fixes and contributions of the previous versions.
• Support for building with Indigo has been removed
• A lot of dead code has been removed
• All palloc calls have been replaced by palloc0
• MOLECULE is not binary compatible and needs to be recreated when you upgrade from 3.2 to 4.0!
• MS spectrum creation has been removed. You may use enviPat instead for much more accurate spectra 

So: CREATE EXTENSION pgchem_tigress;

tcn again

In the previous post, I suggested that contrib/tcn for example could be used to selectively update caches.

OK, here is my try:


By touching the cache only when the database signals a change of the base table and using the passed information to touch the affected rows only, this should be much more efficient than other strategies like routinely invalidating the cache every n minutes or so.