Is there a better approach in SQL to group actions by day/week/month/year/alltime?-CodePudding

I'm thinking my way to do that is a little archaic, not optimized... i don't need super detailed statistics, lets say i want the number of clicks on a link on a blog post per (actual) day/week/month/year nothing more, don't want the hour of click, just a number for each corresponding times (day/month/year).

I've created this table :

CREATE TABLE `clicks` (
  `file_id` bigint(20) unsigned NOT NULL,
  `filename` varchar(100) NOT NULL,
  `day` int(10) unsigned DEFAULT 0,
  `week` int(10) unsigned DEFAULT 0,
  `month` int(10) unsigned DEFAULT 0,
  `year` int(10) unsigned DEFAULT 0,
  `all` int(10) unsigned DEFAULT 0,
  PRIMARY KEY (`file_id`)
)

And each time there's a click, i update every column of a row by 1.

UPDATE clicks SET day = day 1, week = week 1 [..] WHERE file_id = $id

And at every end of day/week/month/year there's a cronjob who will reset the corresponding column for every file. For each day end it will be :

UPDATE clicks SET day = 0 [No WHERE Clause]

And when there's new click on a file tomorrow, it'll increment the day column again.

I have a small VPS server, small storage space, small RAM etc.. i just need how many times a file has been clicked this day only (not yesterday), this week (not the week before) etc.. and i'm trying to not have big & slow queries by having a line for each click and having millions of them.

Is my way of doing seems ok, or is there a better approach for what i want ?

Thanks everyone for the help.

CodePudding user response：

You could create a table just storing the clicks, something like this:

CREATE TABLE clicks (
    file_id INT NOT NULL,
    filename VARCHAR(100) NOT NULL,
    click_time TIMESTAMP NOT NULL
);

Then you just need to use the group by, to extract the clicks. For Example:

-- group clicks by day
SELECT DATE(click_time) AS day, COUNT(*) AS clicks
FROM clicks
GROUP BY day;

-- group clicks by week
SELECT YEARWEEK(click_time) AS week, COUNT(*) AS clicks
FROM clicks
GROUP BY week;

This is quite more efficient and requires less storage

CodePudding user response：

Build and maintain a "Summary table" by date. Only summarize by DAY, then sum up the counts to get "by week", etc. That also lets you get the tallies for arbitrary ranges of days.