Pourquoi mes données PostgreSQL de Discourse sont-elles trop volumineuses ?

Soutien Installation
1

Today I was evaluating my discourse server and found two weird things.

1- If I understand correctly, the discourse postgres_data folder which is in /var/discourse/shared/standalone folder, is where discourse stores the database. Now this folder for my forum is about 8GB. Which I believe is too big for a humble forum. Can someone explain why its too big?

2- I have another folder named postgres_data_old that is also about 7GB. What is this for?

Also, my server memory was about 4GB. I found its mostly consumed. So I upgraded it to 8GB. Again I think a humble forum shouldn’t need that much of memory.

2

There was a postgres update recently, there is a old folder in case something went wrong.

If your forum works fine, you can execute this command

cd /var/discourse
./launcher cleanup app

It should clean the old postgres data folder.

For the memory, DIscourse works this way, it uses the most memory it can, you don’t have to worry about that.

If you want it to use less memory, you can change the db_shared_buffers in the app.yml (command : nano containers/app.yml)

6 « J'aime »
3

what about the big database? Is it something wrong with my forum? I think its way too larger than it should be.

4

How many posts do you have ?

I’ll let a specialist give you an answer about that. I know that Discourse stores a lot of information to provide relevent statistics and a good search engine I guess. It may not be alarming.

5

You can run the following commands to see which tables are taking up the most disk space

./launcher enter app
su - postgres
psql discourse

SELECT nspname || '.' || relname AS "relation",
    pg_size_pretty(pg_total_relation_size(C.oid)) AS "total_size"
  FROM pg_class C
  LEFT JOIN pg_namespace N ON (N.oid = C.relnamespace)
  WHERE nspname NOT IN ('pg_catalog', 'information_schema')
    AND C.relkind <> 'i'
    AND nspname !~ '^pg_toast'
  ORDER BY pg_total_relation_size(C.oid) DESC
  LIMIT 20;
5 « J'aime »
6

This is the output of my forum for these commands. I think the first 5 rows are consuming too much space. I cant imagine why user_actions should be around 2GB. Or post timing about 1GB. Can you give me an idea what could be wrong?

On the other hand, is there any way to clean unnecessary data? For example maybe I can get rid of most of email_logs. I don’t send too many emails. I don’t know why this is so big.

 public.user_actions     | 1792 MB
 public.email_logs       | 1293 MB
 public.post_timings     | 731 MB
 public.directory_items  | 456 MB
 public.topic_views      | 446 MB
 public.posts            | 380 MB
 public.post_search_data | 298 MB
 public.notifications    | 170 MB
 backup.topic_views      | 156 MB
 public.post_actions     | 155 MB
 public.user_histories   | 134 MB
 backup.directory_items  | 123 MB
 public.user_auth_tokens | 110 MB
 public.users            | 91 MB
 backup.user_auth_tokens | 87 MB
 public.user_visits      | 77 MB
 backup.posts            | 68 MB
 backup.post_timings     | 65 MB
 backup.post_search_data | 64 MB
 public.optimized_images | 63 MB
1 « J'aime »
7

Did you do an upgrade recently?

I’ve seen a commit about cleaning the email logs table. After the rebuild, it may be lighter

Also, you might check the setting delete email logs after days, it should be safer than deleting it manually

3 « J'aime »
8

Thank you very much. I found this option and changed it from 90 days to 10 days.

What about user_actions ? what is stored in this table that it has got so big? I have the same question about post_timings and topic_views too. The names indicate that these should be just a bunch of numbers. And shouldn’t take really this much space.

1 « J'aime »
9

You can run a backup, download it, and analyze it locally. A pg_dump is just a text file that is human readable and will let you check what exactly is in each table.

1 « J'aime »
10

I followed your suggestion and downloaded and extracted the backup. it was about 2GB. Is it normal that its 1/4 of what discourse reports?

Btw I realized that a huge amount of data is for excessive number of inactive users. Its more than 100k. Is there an automatic way for deleting all these users? they don’t have any post or other things that might break the process.

If there is no automatic way, If I remove them with api call, does it also clear all information related to them from database?

11

2GB was the size of the compressed backup? Also backups don’t include indexes, and they take a lot of space.

An inactive user without posts or likes is just a single line in the users table. Are you sure all the space is coming from inactive users?

Did your forum came in from an import? Maybe the import created some bad data?

2 « J'aime »
12

No. Its the size of extracted backup. The size of compressed backup is about 300MB

13 
backup.topic_views

I looks like you did a restore recently and the data before the restore is backup-ed in the backup schema. If you’re certain that you no longer need to recover back to the previous state, you can drop the schema by running the following commands.

./launcher enter app
su - postgres
psql discourse
ALTER SCHEMA "backup" TO "backup-moved";
# Check that you site is still working and up to date
DROP SCHEMA "backup-moved" CASCADE;
6 « J'aime »
14

J’ai pensé à raviver ce sujet, car il s’agit du même problème.

Notre dossier postgres_data de Discourse fait 75 Go, ce qui me semble énorme. Selon le panneau d’administration, une sauvegarde fait environ 10,5 Go et les uploads représentent environ 9,3 Go.

J’ai vérifié quelles tables occupaient le plus d’espace et voici ce que j’ai obtenu :

 public.posts                | 51 Go
 public.post_search_data     | 9769 Mo
 public.post_timings         | 3997 Mo
 public.user_actions         | 2144 Mo
 public.post_custom_fields   | 1039 Mo
 public.topics               | 676 Mo
 public.post_stats           | 663 Mo
 public.post_replies         | 643 Mo
 public.quoted_posts         | 523 Mo
 public.user_visits          | 476 Mo
 public.top_topics           | 403 Mo
 public.user_auth_token_logs | 364 Mo
 public.topic_links          | 353 Mo
 public.topic_users          | 335 Mo
 public.topic_views          | 301 Mo
 public.user_histories       | 220 Mo
 public.users                | 209 Mo
 public.stylesheet_cache     | 194 Mo
 public.directory_items      | 143 Mo
 public.notifications        | 139 Mo

Je me demande s’il est normal que la table public.posts prenne autant de place (51 Go). Nous parlons d’un forum avec 6 millions de posts, ce qui ne me semble pas extraordinaire.

Est-ce normal ?

3 « J'aime »
15

Compléter les informations ci-dessus avec rake db:stats :

table_name | row_estimate | table_size | index_size | total_size

posts | 8847417 | 39 Go | 12 Go | 51 Go
post_search_data | 5880635 | 8377 Mo | 1392 Mo | 9769 Mo
post_timings | 23728606 | 1571 Mo | 2430 Mo | 4001 Mo
user_actions | 5424982 | 488 Mo | 1657 Mo | 2144 Mo
post_custom_fields | 5832468 | 429 Mo | 609 Mo | 1039 Mo

J’ai vu d’autres exemples où 10 millions de lignes pour la table posts se traduisent par environ 15 Go. Nous avons maintenant 8 millions de lignes avec une taille de 39 Go.

Y a-t-il un moyen d’optimiser cela ?

16

Il y a beaucoup de facteurs en jeu ici. Par exemple, les longs messages occupent plus d’espace que les courts. En tant que lusophone moi aussi, je sais à quel point notre langue peut être verbeuse. Je vois que vos données proviennent d’une importation. Certains artefacts, comme les messages avec des citations imbriquées sur 5 niveaux, ne sont pas courants dans Discourse mais se retrouvent sur votre site à cause de l’importation. Notre langue compte aussi : un ç occupe le double de l’espace d’un s.

Je pense que nous ne modifions pas le paramètre par défaut de PostgreSQL, et que la colonne posts.raw est gérée par TOAST et compressée.

5 « J'aime »
17

Merci, cela a fonctionné. Je pense qu’il manque RENAME dans cette ligne. Je l’ai fait ainsi :

ALTER SCHEMA "backup" RENAME TO "backup-moved";
2 « J'aime »