I have a query that gets all the info I need for a messaging system's main page (including unread message count, etc)... but it currently retrieves the original threads message. I would like to augment the below query to grab the most recent message in each thread instead.
This query is very close, however my mediocre SQL skills are keeping me from wrapping things up...
$messages = array();
$unread_messages_total = 0;
$messages_query = "
SELECT m.*
, COUNT(r.id) AS num_replies
, MAX(r.datetime) AS reply_datetime
, (m.archived NOT LIKE '%,".$cms_user['id'].",%') AS message_archive
, (m.viewed LIKE '%,".$cms_user['id'].",%') AS message_viewed
, SUM(r.viewed NOT LIKE '%,".$cms_user['id'].",%') AS unread_replies
, CASE
WHEN MAX(r.datetime) >= m.datetime THEN MAX(r.datetime)
ELSE m.datetime
END AS last_datetime
FROM directus_messages AS m
LEFT JOIN directus_messages as r ON m.id = r.reply
WHERE m.active = '1'
AND (m.to LIKE '%,".$cms_user['id'].",%' OR m.to = 'all' OR m.from = '".$cms_user['id']."')
GROUP BY m.id
HAVING m.reply = '0'
ORDER BY last_datetime DESC";
foreach($dbh->query($messages_query) as $row_messages){
$messages[] = $row_messages;
$unread_messages_total += (strpos($row_messages['archived'], ','.$cms_user['id'].',') === false && ( (strpos($row_messages['viewed'], ','.$cms_user['id'].',') === false && $row_messages['unread_replies'] == NULL) || ($row_messages['unread_replies']>0 && $row_messages['unread_replies'] != NULL) ) )? 1 : 0;
}
Thanks in advance for any help you can provide!
EDIT: (Database)
CREATE TABLE `cms_messages` (
`id` int(10) NOT NULL auto_increment,
`active` tinyint(1) NOT NULL default '1',
`subject` varchar(255) NOT NULL default '',
`message` text NOT NULL,
`datetime` datetime NOT NULL default '0000-00-00 00:00:00',
`reply` int(10) NOT NULL default '0',
`from` int(10) NOT NULL default '0',
`to` varchar(255) NOT NULL default '',
`viewed` varchar(255) NOT NULL default ',',
`archived` varchar(255) NOT NULL default ',',
PRIMARY KEY (`id`)
) ENGINE=MyISAM AUTO_INCREMENT=1 DEFAULT CHARSET=latin1;
EDIT 2: (Requirements)
Return all parent messages for a specific user_id: $cms_user['id']
Return the number of replies for that parent message: num_replies
Return the number of unread replies for that parent message: unread_replies
Return the date of the parent message or it's most recent reply: last_datetime
Return whether the message is in the archive: message_archive
Return whether the message has been viewed: message_viewed
Return all messages in DESC datetime order
Return the newest message, from the parent or replies if there are some (like gmail)
If you have only 2 levels of messages (i.e., only parent messages and direct answers), you might try this query:
select
root_message.id,
root_message.active,
root_message.subject,
case
when max_reply_id.max_id is null then
root_message.message
else
reply_message.message
end as message,
root_message.datetime,
root_message.reply,
root_message.from,
root_message.to,
root_message.viewed,
root_message.archived
from
-- basic data
cms_messages as root_message
-- ID of last reply for every root message
left join (
select
max(id) as max_id,
reply as parent_id
from
cms_messages
where
reply <> 0
group by
reply
) as max_reply_id on max_reply_id.parent_id = root_message.id
left join cms_messages as reply_message on reply_message.id = max_reply_id.max_id
where
root_message.reply = 0
It uses subquery max_reply_id as source of data to select ID of the latest answer. If it exists (i.e., if there are answers), reply_message.message is used. If it does not exist (no answer has been found for root message), then root_message.message is used.
You should also think about structure of table. E.g., it would make more sense if reply contained either NULL, if it is parent message, or ID of existing message. Currently, you set it to 0 (ID of non-existent message), which is wrong. Types of viewed and archived are also weird.
Edit: you should also avoid using having clause. Use where instead, when possible.
Here's a new query that should fulfil your requirements. If there is any problem with it (i.e., if it returns wrong data), let me know.
Like the first query, it:
uses subquery reply_summary to accumulate data about replies (ID of last reply, number of replies and number of unread replies);
joins this subquery to the base table;
joins cms_messages as reply_message to the subquery, based on reply_summary.max_reply_id, to get data about the last reply (message, datetime).
I've simplified the way how you determine last_datetime - it now takes either time of last reply (if there is any reply), or time of original post (when no replies are found).
I have not filtered replies by from and to fields. If it is necessary, where clause of reply_summary subquery should be updated.
select
parent_message.id,
parent_message.subject,
parent_message.message,
parent_message.from,
parent_message.to,
coalesce(reply_summary.num_replies, 0) as num_replies,
last_reply_message.datetime as reply_datetime,
(parent_message.archived NOT LIKE '%,{$cms_user['id']},%') AS message_archive,
(parent_message.viewed LIKE '%,{$cms_user['id']},%') AS message_viewed,
reply_summary.unread_replies,
coalesce(last_reply_message.message, parent_message.message) as last_message,
coalesce(last_reply_message.datetime, parent_message.datetime) as last_datetime
from
cms_messages as parent_message
left join (
select
reply as parent_id,
max(id) as last_reply_id,
count(*) as num_replies,
sum(viewed not like '%,{$cms_user['id']},%') as unread_replies
from
cms_messages
where
reply <> 0 and
active = 1
group by
reply
) as reply_summary on reply_summary.parent_id = parent_message.id
left join cms_messages as last_reply_message on last_reply_message.id = reply_summary.last_reply_id
where
parent_message.reply = 0 and
parent_message.active = 1 and
(parent_message.to like '%,{$cms_user['id']},%' or parent_message.to = 'all' or parent_message.from = '{$cms_user['id']}')
order by
last_datetime desc;
your problem is that you are fetching only m records no matter what the order of the r records.
try adding
SELECT m.*, r.*
or
SELECT r.*, m.*
if you are using PDO::FETCH_ASSOC as your PDO fetch mode (assuming you are using PDO to access your database), the result will be an associative array where if the result set contains multiple columns with the same name, PDO::FETCH_ASSOC returns only a single value per column name. not sure which order takes presidence, so you would have to try both.
if your columns are defined in the right order, they will return the r.* value if one exists, or the m.* value if no r records exist. does this make sense? this way your result set will contain the latest record no matter which table (m or r) contains them.
http://www.php.net/manual/en/pdo.constants.php
I am afraid that you wont be able to solve this problem with a single query. Either you have to use more queries and gather the informations in the surrounding code or you will have to redesign the database structure for your messaging system a litte (tables: threads, posts, etc.). If you decide to redesign the database structure, you should also take care of the way you handle the viewed and archived fields. The way you use the fields (varchar 255 only!) might work for some users, but as soon as there are more users and higher user IDs your message system will break down.
Related
I have 3 tables course, grade and evaluation. I want comparing two tables grade and evaluation . if the data in the table grade does not exist in the table evaluation , then the data will appear (output)
" select Grade.ID_Courses,Course.ID_Courses,Grade.NAME,
Course.NAME, Grade.ID_Courses,
Evaluation.NAME,
Evaluation.Year,
Grade.Year
from Grade, Course, Evaluation
WHERE
Grade.ID_Courses=Course.ID_Courses AND
Grade.NAME=JOHN and
Grade.Year=1 and
Evaluation.NAME=GRADE.NAME and
Grade.ID_Courses NOT IN (SELECT ID_Courses FROM Evaluation where NAME=JOHN and Year=1 )
GROUP BY Grade.ID_Courses"
the problem is when the name john is not in the table evaluation then there is no output comes out .
Avoid NOT IN like the plague if
SELECT ID_Courses FROM Evaluation where `NAME`='JOHN' and Year=1
could ever contain NULL. Instead, use NOT EXISTS or Left Joins
use explicit joins, not 1980's style joins using the WHERE clause
To illustrate the misery of NOT IN:
SQL NOT IN () danger
create table mStatus
( id int auto_increment primary key,
status varchar(10) not null
);
insert mStatus (status) values ('single'),('married'),('divorced'),('widow');
create table people
( id int auto_increment primary key,
fullName varchar(100) not null,
status varchar(10) null
);
Chunk1:
truncate table people;
insert people (fullName,`status`) values ('John Henry','single');
select * from mstatus where `status` not in (select status from people);
** 3 rows, as expected **
Chunk2:
truncate table people;
insert people (fullName,`status`) values ('John Henry','single'),('Kim Billings',null);
select * from mstatus where status not in (select status from people);
no rows, huh?
Obviously this is 'incorrect'. It arises from SQL's use of three-valued logic,
driven by the existence of NULL, a non-value indicating missing (or UNKNOWN) information.
With NOT IN, Chunk2 it is translated like this:
status NOT IN ('married', 'divorced', 'widowed', NULL)
This is equivalent to:
NOT(status='single' OR status='married' OR status='widowed' OR status=NULL)
The expression "status=NULL" evaluates to UNKNOWN and, according to the rules of three-valued logic,
NOT UNKNOWN also evaluates to UNKNOWN. As a result, all rows are filtered out and the query returns an empty set.
Possible solutions include:
select s.status
from mstatus s
left join people p
on p.status=s.status
where p.status is null
or use not exists
Try using joins to solve this
select g.*, e.*,c.* from
grade g inner join evaluation e on
g.ID_COURSES <> e.ID_COURSES and g.year <> e.year
inner join COURSE c on c.ID_COURSES = g.ID_COURSES
;
The requirement is to find the first available identifier where an identifier is an alphanumeric string, such as:
ABC10000
ABC10345
ABC88942
ABC90123
The database table has a structure such as:
id, user, identifier
Note that the alpha component ABC is consistent throughout and won't change. The numeric component should be between 10000 and 99999.
How best to tackle this? It does not seem like an overly complex problem - looking for simplest solution using either just MySQL or a combination of SQL and PHP. The current solution pulls each record from the database into an array and then loops from 10000 onwards, prepending ABC and checking availability, which seems like it could be improved significantly.
Edit: Original question was not clear enough in that a certain amount of identifiers have been assigned already, and I am looking to fill in the gaps. From the short list I provided, the next available would be ABC10001. Eventually, however, it would be ABC10346 and then ABC88943 and so on
Edit: Sorry for a poorly structured question. To avoid any further confusion, here is the actual table structure:
CREATE TABLE IF NOT EXISTS `User_Loc` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`user` int(11) DEFAULT NULL,
`value` varchar(100) COLLATE utf8_unicode_ci DEFAULT NULL,
PRIMARY KEY (`id`),
UNIQUE KEY `UNIQ_64FB41DA17323CBC` (`user`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci AUTO_INCREMENT=4028 ;
You have to self join the table and look for the first NULL value in the joined table
SELECT CONCAT('ABC', SUBSTRING(t1.value, 4)+1) AS next_value
FROM test t1
LEFT JOIN test t2 on SUBSTRING(t1.value, 4)+1 = SUBSTRING(t2.value, 4)
WHERE ISNULL(t2.value)
ORDER BY t1.value ASC
LIMIT 1
http://sqlfiddle.com/#!2/d69105/22
edit
With the comment about some 'specialities' at ncatnow. There are slight adjusments to make with the help of subselects for ridding the 'ABC' and UNION for having a default value
SELECT
CONCAT('ABC', t1.value+1) AS next_value
FROM
((SELECT '09999' AS value) UNION (SELECT SUBSTRING(value, 4) AS value FROM test)) t1
LEFT JOIN
((SELECT '09999' AS value) UNION (SELECT SUBSTRING(value, 4) AS value FROM test)) t2
ON t1.value+1 = t2.value
WHERE
ISNULL(t2.value)
AND t1.value >= '09999'
ORDER BY t1.value ASC
LIMIT 1
http://sqlfiddle.com/#!2/28acf6/50
Similar to the above reply by #HerrSerker, but this will cope with existing identifiers which have the numeric part starting with a zero.
SELECT CONCAT('ABC',SUBSTRING(CONCAT('00000', CAST((CAST(SUBSTRING(a.identifier, 4) AS SIGNED) + 1) AS CHAR)), -5)) AS NextVal
FROM SomeTable a
LEFT OUTER JOIN SomeTable b
ON b.identifier = CONCAT('ABC',SUBSTRING(CONCAT('00000', CAST((CAST(SUBSTRING(a.identifier, 4) AS SIGNED) + 1) AS CHAR)), -5))
WHERE b.identifier IS NULL
ORDER BY NextVal
LIMIT 1
what comes to my mind is one table with all indentifiers and use this sql
SELECT identifier FROM allIdentifiersTable WHERE identifier NOT IN (SELECT identifier FROM yourTable) LIMIT 1
Reconsidering this from your edit, you should go the PHP route and add another table or other means to store the last filled id:
$identifier = 0;
$i = mysql_query("SELECT identifier FROM last_identifier");
if ($row = mysql_fetch_assoc($i)) $identifier = $row["identifier"];
if ($identifier < 10000) $identifier = 10000;
do {
$identifier += 1;
$result = mysql_query("
INSERT IGNORE INTO table (id, user, identifier)
VALUES ('[...]', '[...]',
'ABC" . str_pad($identifier, 5, "0", STR_PAD_LEFT) . "'
)");
if (mysql_affected_rows($result) < 1) continue;
} while (false);
mysql_query("UPDATE last_identifier SET identifier = '$identifier'");
Of course, you need to add a UNIQUE index on the identifier field.
I have 3 queries. I was told that they were potentially inefficient so I was wondering if anyone who is experienced could suggest anything. The logic is somewhat complex so bear with me.
I have two tables: shoutbox, and topic. Topic stores all information on topics that were created, while shoutbox stores all comments pertaining to each topic. Each comment comes with a group labelled by reply_chunk_id. The earliest timestamp is the first comment, while any following with the same reply_chunk_id and a later timestamp are replies. I would like to find the latest comment for each group that was started by the user (made first comment) and if the latest comment was made this month display it.
What I have written achieves that with one problem: all the latest comments are displayed in random order. I would like to organize these groups/latest comments. I really appreciate any advice
Shoutbox
Field Type
-------------------
id int(5)
timestamp int(11)
user varchar(25)
message varchar(2000)
topic_id varchar(35)
reply_chunk_id varchar(35)
Topic
id mediumint(8)
topic_id varchar(35)
subject_id mediumint(8)
file_name varchar(35)
topic_title varchar(255)
creator varchar(25)
topic_host varchar(255)
timestamp int(11)
color varchar(10)
mp3 varchar(75)
custom_background varchar(55)
description mediumtext
content_type tinyint(1)
Query
$sql="SELECT reply_chunk_id FROM shoutbox
GROUP BY reply_chunk_id
HAVING count(*) > 1
ORDER BY timestamp DESC ";
$stmt16 = $conn->prepare($sql);
$result=$stmt16->execute();
while($row = $stmt16->fetch(PDO::FETCH_ASSOC)){
$sql="SELECT user,reply_chunk_id, MIN(timestamp) AS grp_timestamp
FROM shoutbox WHERE reply_chunk_id=? AND user=?";
$stmt17 = $conn->prepare($sql);
$result=$stmt17->execute(array($row['reply_chunk_id'],$user));
while($row2 = $stmt17->fetch(PDO::FETCH_ASSOC)){
$sql="SELECT t.topic_title, t.content_type, t.subject_id,
t.creator, t.description, t.topic_host,
c1.message, c1.topic_id, c1.user, c1.timestamp AS max
FROM shoutbox c1
JOIN topic t ON (t.topic_id = c1.topic_id)
WHERE reply_chunk_id = ? AND c1.timestamp > ?
ORDER BY c1.timestamp DESC, c1.id
LIMIT 1";
$stmt18 = $conn->prepare($sql);
$result=$stmt18->execute(array($row2['reply_chunk_id'],$month));
while($row3 = $stmt18->fetch(PDO::FETCH_ASSOC)){
Make the first query:
SELECT reply_chunk_id FROM shoutbox
GROUP BY reply_chunk_id
HAVING count(*) > 1
ORDER BY timestamp DESC
This does the same, but is faster.
Make sure you have an index on reply_chunk_id.
The second query:
SELECT user,reply_chunk_id, MIN(timestamp) AS grp_timestamp
FROM shoutbox WHERE reply_chunk_id=? AND user=?
The GROUP BY is unneeded, because only one row gets returned, because of the MIN() and the equality tests.
The third query:
SELECT t.topic_title, t.content_type, t.subject_id,
t.creator, t.description, t.topic_host,
c1.message, c1.topic_id, c1.user, c1.timestamp AS max
FROM shoutbox c1
JOIN topic t ON (t.topic_id = c1.topic_id)
WHERE reply_chunk_id = ? AND c1.timestamp > ?
ORDER BY c1.timestamp DESC, c1.id
LIMIT 1
Doing it all in one query:
SELECT
t.user,t.reply_chunk_id, MIN(t.timestamp) AS grp_timestamp,
t.topic_title, t.content_type, t.subject_id,
t.creator, t.description, t.topic_host,
c1.message, c1.topic_id, c1.user, c1.timestamp AS max
FROM shoutbox c1
INNER JOIN topic t ON (t.topic_id = c1.topic_id)
LEFT JOIN shoutbox c2 ON (c1.id = c2.id and c1.timestamp < c2.timestamp)
WHERE c2.timestamp IS NULL AND t.user = ?
GROUP BY t.reply_chunk_id
HAVING count(*) > 1
ORDER BY t.reply_chunk_id
or the equivalent
SELECT
t.user,t.reply_chunk_id, MIN(t.timestamp) AS grp_timestamp,
t.topic_title, t.content_type, t.subject_id,
t.creator, t.description, t.topic_host,
c1.message, c1.topic_id, c1.user, c1.timestamp AS max
FROM shoutbox c1
INNER JOIN topic t ON (t.topic_id = c1.topic_id)
WHERE c1.timestamp = (SELECT max(timestamp) FROM shoutbox c2
WHERE c2.reply_chunk_id = c1.reply_chunk_id)
AND t.user = ?
GROUP BY t.reply_chunk_id
HAVING count(*) > 1
ORDER BY t.reply_chunk_id
How does this work?
The group by selects one entry per topic.reply_chunk_id
The left join (c1.id = c2.id and c1.`timestamp` < c2.`timestamp`) + WHERE c2.`timestamp` IS NULL selects only those items from shoutbox which have the highest timestamp. This works because MySQL keeps increasing c1.timestamp to get c2.timestamp to be null as soon as that is true, it c1.timestamp will have reached its maximum value and will select that row within the possible rows to choose from.
If you don't understand point 2, see: http://dev.mysql.com/doc/refman/5.0/en/example-maximum-column-group-row.html
Note that the PDO is autoescaping the fields with backticks
Sounds like most of it should be directly from your ShoutBox table. Prequery to find all "Chunks" the user replied to... of those chunks (and topic_ID since each chunk is always the same topic), get their respective minimum and maximum. Using the "Having count(*) > 1" will force only those that HAVE a second posting by a given user (what you were looking for).
THEN, re-query to the chunks to get the minimum regardless of user. This prevents the need of querying ALL chunks. Then join only what a single user is associated with back to the Topic.
Additionally, and I could be incorrect and need to adjust (minimally), but it appears that the SOUNDBOX table ID column would be an auto-increment column, and just happens to be time-stamped too at time of creation. That said, for a given "Chunk", the earliest ID would be the same as the earliest timestamp as they would be stamped at the same time they are created. Also makes easier on subsequent JOINs and sub query too.
By using STRAIGHT_JOIN, should force the "PreQuery" FIRST, come up with a very limited set, then qualify the WHERE clause and joins afterwords.
select STRAIGHT_JOIN
T.topic_title,
T.content_type,
T.subject_id,
T.creator,
T.description,
T.topic_host,
sb2.Topic_ID
sb2.message,
sb2.user,
sb2.TimeStamp
from
( select
sb1.Reply_Chunk_ID,
sb1.Topic_ID,
count(*) as TotalEntries,
min( sb1.id ) as FirstIDByChunkByUser,
min( sbJoin.id ) as FirstIDByChunk,
max( sbJoin.id ) as LastIDByChunk,
max( sbJoin.timestamp ) as LastTimeByChunk
from
ShoutBox sb1
join ShoutBox sbJoin
on sb1.Reply_Chunk_ID = sbJoin.Reply_Chunk_ID
where
sb1.user = CurrentUser
group by
sb1.Reply_Chunk_ID,
sb1.Topic_ID
having
min( sb1.id ) = min( sbJoin.ID ) ) PreQuery
join Topic T on
PreQuery.Topic_ID = T.ID
join ShoutBox sb2
PreQuery.LastIDByChunk = sb2.ID
where
sb2.TimeStamp >= YourTimeStampCriteria
order by
sb2.TimeStamp desc
EDIT ---- QUERY EXPLANATION -- with Modified query.
I've changed the query from re-reading (as was almost midnight when answered after holiday weekend :)
First, "STRAIGHT_JOIN" is a MySQL clause telling the engine to "do the query in the way / sequence I've stated". Basically, sometimes an engine will try to think for you and optimize in ways that may appear more efficient, but if based on your data, you know what will retrieve the smallest set of data first, and then join to other lookup fields next might in fact be better. Second the "PreQuery". If you have a "SQL-Select" statement (within parens) as Alias "From" clause, The "PreQuery" is just the name of the alias of the resultset... I could have called it anything, just makes sense that this is a stand-alone query of it's own. (Ooops... fixed to ShoutBox :) As for case-sensitivity, typically Column names are NOT case-sensitive... However, table names are... You could have a table name "MyTest" different than "mytest" or "MYTEST". But by supplying "alias", it helps shorten readability (especially with VeryLongTableNamesUsed ).
Should be working after the re-reading and applying adjustments.. Try the first "Prequery" on its own to see how many records it returns. On its own merits, it should return... for a single "CurrentUser" parameter value, every "Reply_Chunk_ID" (which will always have the same topic_id", get the first ID the person entered (min()). By JOINing again to Shoutbox on the chunk id, we (only those qualified as entered by the user), get the minimum and maximum ID per the chunk REGARDLESS of who started or responded. By applying the HAVING clause, this should only return those where the same person STARTED the topic (hence both have the same min() value.)
Finally, once those have been qualified, join directly to the TOPIC and SHOUTBOX tables again on their own merits of topic_id and LastIDByChunk and order the final results by the latest comment response timestamp descending.
I've added a where clause to further limit your "timestamp" criteria where the most recent final timestamp is on/after the given time period you want.
I would be curious how this query's time performance works compared to your already accepted answer too.
This post is taking a substantial amount of time to type because I'm trying to be as clear as possible, so please bear with me if it is still unclear.
Basically, what I have are a table of posts in the database which users can add privacy settings to.
ID | owner_id | post | other_info | privacy_level (int value)
From there, users can add their privacy details, allowing it to be viewable by all [privacy_level = 0), friends (privacy_level = 1), no one (privacy_level = 3), or specific people or filters (privacy_level = 4). For privacy levels specifying specific people (4), the query will reference the table "post_privacy_includes_for" in a subquery to see if the user (or a filter the user belongs to) exists in a row in the table.
ID | post_id | user_id | list_id
Also, the user has the ability to prevent some people from viewing their post in within a larger group by excluding them (e.g., Having it set for everyone to view but hiding it from a stalker user). For this, another reference table is added, "post_privacy_exclude_from" - it looks identical to the setup as "post_privacy_includes_for".
My problem is that this does not scale. At all. At the moment, there are about 1-2 million posts, the majority of them set to be viewable by everyone. For each post on the page it must check to see if there is a row that is excluding the post from being shown to the user - this moves really slow on a page that can be filled with 100-200 posts. It can take up to 2-4 seconds, especially when additional constraints are added to the query.
This also creates extremely large and complex queries that are just... awkward.
SELECT t.*
FROM posts t
WHERE ( (t.privacy_level = 3
AND t.owner_id = ?)
OR (t.privacy_level = 4
AND EXISTS
( SELECT i.id
FROM PostPrivacyIncludeFor i
WHERE i.user_id = ?
AND i.thought_id = t.id)
OR t.privacy_level = 4
AND t.owner_id = ?)
OR (t.privacy_level = 4
AND EXISTS
(SELECT i2.id
FROM PostPrivacyIncludeFor i2
WHERE i2.thought_id = t.id
AND EXISTS
(SELECT r.id
FROM FriendFilterIds r
WHERE r.list_id = i2.list_id
AND r.friend_id = ?))
OR t.privacy_level = 4
AND t.owner_id = ?)
OR (t.privacy_level = 1
AND EXISTS
(SELECT G.id
FROM Following G
WHERE follower_id = t.owner_id
AND following_id = ?
AND friend = 1)
OR t.privacy_level = 1
AND t.owner_id = ?)
OR (NOT EXISTS
(SELECT e.id
FROM PostPrivacyExcludeFrom e
WHERE e.thought_id = t.id
AND e.user_id = ?
AND NOT EXISTS
(SELECT e2.id
FROM PostPrivacyExcludeFrom e2
WHERE e2.thought_id = t.id
AND EXISTS
(SELECT l.id
FROM FriendFilterIds l
WHERE l.list_id = e2.list_id
AND l.friend_id = ?)))
AND t.privacy_level IN (0, 1, 4))
AND t.owner_id = ?
ORDER BY t.created_at LIMIT 100
(mock up query, similar to the query I use now in Doctrine ORM. It's a mess, but you get what I am saying.)
I guess my question is, how would you approach this situation to optimize it? Is there a better way to set up my database? I'm willing to completely scrap the method I have currently built up, but I wouldn't know what to move onto.
Thanks guys.
Updated: Fix the query to reflect the values I defined for privacy level above (I forgot to update it because I simplified the values)
Your query is too long to give a definitive solution for, but the approach I would follow is to simply the data lookups by converting the sub-queries into joins, and then build the logic into the where clause and column list of the select statement:
select t.*, i.*, r.*, G.*, e.* from posts t
left join PostPrivacyIncludeFor i on i.user_id = ? and i.thought_id = t.id
left join FriendFilterIds r on r.list_id = i.list_id and r.friend_id = ?
left join Following G on follower_id = t.owner_id and G.following_id = ? and G.friend=1
left join PostPrivacyExcludeFrom e on e.thought_id = t.id and e.user_id = ?
(This might need expanding: I couldn't follow the logic of the final clause.)
If you can get the simple select working fast AND including all the information needed, then all you need to do is build up the logic in the select list and where clause.
Had a quick stab at simplifying this without re-working your original design too much.
Using this solution your web page can now simply call the following stored procedure to get a list of filtered posts for a given user within a specified period.
call list_user_filtered_posts( <user_id>, <day_interval> );
The whole script can be found here : http://pastie.org/1212812
I haven't fully tested all of this and you may find this solution isn't performant enough for your needs but it may help you in fine tuning/modifying your existing design.
Tables
Dropped your post_privacy_exclude_from table and added a user_stalkers table which works pretty much like the inverse of user_friends. Kept the original post_privacy_includes_for table as per your design as this allows a user restrict a specific post to a subset of people.
drop table if exists users;
create table users
(
user_id int unsigned not null auto_increment primary key,
username varbinary(32) unique not null
)
engine=innodb;
drop table if exists user_friends;
create table user_friends
(
user_id int unsigned not null,
friend_user_id int unsigned not null,
primary key (user_id, friend_user_id)
)
engine=innodb;
drop table if exists user_stalkers;
create table user_stalkers
(
user_id int unsigned not null,
stalker_user_id int unsigned not null,
primary key (user_id, stalker_user_id)
)
engine=innodb;
drop table if exists posts;
create table posts
(
post_id int unsigned not null auto_increment primary key,
user_id int unsigned not null,
privacy_level tinyint unsigned not null default 0,
post_date datetime not null,
key user_idx(user_id),
key post_date_user_idx(post_date, user_id)
)
engine=innodb;
drop table if exists post_privacy_includes_for;
create table post_privacy_includes_for
(
post_id int unsigned not null,
user_id int unsigned not null,
primary key (post_id, user_id)
)
engine=innodb;
Stored Procedures
The stored procedure is relatively simple - it initially selects ALL posts within the specified period and then filters out posts as per your original requirements. I have not performance tested this sproc with large volumes but as the initial selection is relatively small it should be performant enough as well as simplifying your application/middle tier code.
drop procedure if exists list_user_filtered_posts;
delimiter #
create procedure list_user_filtered_posts
(
in p_user_id int unsigned,
in p_day_interval tinyint unsigned
)
proc_main:begin
drop temporary table if exists tmp_posts;
drop temporary table if exists tmp_priv_posts;
-- select ALL posts in the required date range (or whatever selection criteria you require)
create temporary table tmp_posts engine=memory
select
p.post_id, p.user_id, p.privacy_level, 0 as deleted
from
posts p
where
p.post_date between now() - interval p_day_interval day and now()
order by
p.user_id;
-- purge stalker posts (0,1,3,4)
update tmp_posts
inner join user_stalkers us on us.user_id = tmp_posts.user_id and us.stalker_user_id = p_user_id
set
tmp_posts.deleted = 1
where
tmp_posts.user_id != p_user_id;
-- purge other users private posts (3)
update tmp_posts set deleted = 1 where user_id != p_user_id and privacy_level = 3;
-- purge friend only posts (1) i.e where p_user_id is not a friend of the poster
/*
requires another temp table due to mysql temp table problem/bug
http://dev.mysql.com/doc/refman/5.0/en/temporary-table-problems.html
*/
-- the private posts (1) this user can see
create temporary table tmp_priv_posts engine=memory
select
tp.post_id
from
tmp_posts tp
inner join user_friends uf on uf.user_id = tp.user_id and uf.friend_user_id = p_user_id
where
tp.user_id != p_user_id and tp.privacy_level = 1;
-- remove private posts this user cant see
update tmp_posts
left outer join tmp_priv_posts tpp on tmp_posts.post_id = tpp.post_id
set
tmp_posts.deleted = 1
where
tpp.post_id is null and tmp_posts.privacy_level = 1;
-- purge filtered (4)
truncate table tmp_priv_posts; -- reuse tmp table
insert into tmp_priv_posts
select
tp.post_id
from
tmp_posts tp
inner join post_privacy_includes_for ppif on tp.post_id = ppif.post_id and ppif.user_id = p_user_id
where
tp.user_id != p_user_id and tp.privacy_level = 4;
-- remove private posts this user cant see
update tmp_posts
left outer join tmp_priv_posts tpp on tmp_posts.post_id = tpp.post_id
set
tmp_posts.deleted = 1
where
tpp.post_id is null and tmp_posts.privacy_level = 4;
drop temporary table if exists tmp_priv_posts;
-- output filtered posts (display ALL of these on web page)
select
p.*
from
posts p
inner join tmp_posts tp on p.post_id = tp.post_id
where
tp.deleted = 0
order by
p.post_id desc;
-- clean up
drop temporary table if exists tmp_posts;
end proc_main #
delimiter ;
Test Data
Some basic test data.
insert into users (username) values ('f00'),('bar'),('alpha'),('beta'),('gamma'),('omega');
insert into user_friends values
(1,2),(1,3),(1,5),
(2,1),(2,3),(2,4),
(3,1),(3,2),
(4,5),
(5,1),(5,4);
insert into user_stalkers values (4,1);
insert into posts (user_id, privacy_level, post_date) values
-- public (0)
(1,0,now() - interval 8 day),
(1,0,now() - interval 8 day),
(2,0,now() - interval 7 day),
(2,0,now() - interval 7 day),
(3,0,now() - interval 6 day),
(4,0,now() - interval 6 day),
(5,0,now() - interval 5 day),
-- friends only (1)
(1,1,now() - interval 5 day),
(2,1,now() - interval 4 day),
(4,1,now() - interval 4 day),
(5,1,now() - interval 3 day),
-- private (3)
(1,3,now() - interval 3 day),
(2,3,now() - interval 2 day),
(4,3,now() - interval 2 day),
-- filtered (4)
(1,4,now() - interval 1 day),
(4,4,now() - interval 1 day),
(5,4,now());
insert into post_privacy_includes_for values (15,4), (16,1), (17,6);
Testing
As I mentioned before I've not fully tested this but on the surface it seems to be working.
select * from posts;
call list_user_filtered_posts(1,14);
call list_user_filtered_posts(6,14);
call list_user_filtered_posts(1,7);
call list_user_filtered_posts(6,7);
Hope you find some of this of use.
OK, my statement executes well in phpMyAdmin, but not how I expect it in my php page.
This is my statement:
SELECT `egid`, `group_name` , `limit`, MAX( `date` )
FROM employee_groups
GROUP BY `egid`
ORDER BY `egid` DESC ;
This is may table:
CREATE TABLE `employee_groups` (
`egid` int(10) unsigned NOT NULL,
`date` date NOT NULL,
`group_name` varchar(50) NOT NULL,
`limit` smallint(5) unsigned NOT NULL,
PRIMARY KEY (`egid`,`date`)
) ENGINE=MyISAM DEFAULT CHARSET=cp1251;
I want to extract the most recent list of groups, e.g. if a group has been changed I want to have only the last change. And I need it as a list (all groups).
Your query might be broken. You should not select fields that aren't in the group by unless one of the following two conditions apply:
You use an aggregate function.
The value is functionally dependant on the grouped by columns.
The two fields group_name and limit appear to break these rules. This means that you will get indeterminate results for these columns.
If you are trying to select the max per group then you should use a slightly different technique. See Quassnoi's article MYSQL: Selecting records holding a groupwise maximum for a variety of methods you could use.
Here's one way to do it:
SELECT di.*
FROM (
SELECT egid, MAX(date) AS date
FROM employee_groups d
GROUP BY egid
) dd
JOIN employee_groups di
ON di.egid = dd.egid AND di.date = dd.date
aggregate functions will work in mysql, different to the sql standard. to access the value of max(date) from php, you have to alias it:
SELECT `egid`, `group_name` , `limit`, MAX( `date` ) as maxdate
FROM …
you can then select it like any other colum from php with
while($row = mysqli_fetch_row($result)) {
echo htmlspecialchars($row['maxdate']);
}
hope that helps