Restore from s3://path/to/xxx.tgz does not work in docker-compose.yml

[human211]

My docker-compose.yml

version: "3.9"

services:
  mysql-restore:
    image: databack/mysql-backup:1.2.2
    container_name: mysql-restore
    hostname: mysql-restore
    restart: no
    environment:
      DB_SERVER: ${DB_HOST}
      DB_USER: ${DB_USER}
      DB_PASS: ${DB_PASSWORD}
      DB_RESTORE_TARGET: s3://db/db_backup_2025-04-19T19:42:11Z.tgz
      AWS_ACCESS_KEY_ID: ${S3_KEY}
      AWS_SECRET_ACCESS_KEY: ${S3_SECRET_KEY}
      AWS_ENDPOINT_URL: "https://${S3_HOST}/"
      AWS_DEFAULT_REGION: eu-central1
      COMPRESSION: gzip
      DB_DUMP_DEBUG: "true"
    command: restore
    networks:
      - intranet
networks:
  intranet:
    name: intranet
    external: true

I get an error

time="2025-04-19T22:21:58Z" level=fatal msg="requires at least 1 arg(s), only received 0"
mysql-restore exited with code 1

Аlso my attempts to change

DB_RESTORE_TARGET: s3://db/ db_backup_2025-04-19T19:42:11Z.tgz (with space before filename)
DB_RESTORE_TARGET: s3://db/ restore db_backup_2025-04-19T19:42:11Z.tgz

DB_RESTORE_TARGET: s3://db/
with one of
command: restore --target=db_backup_2025-04-19T19:42:11Z.tgz
command: restore --target db_backup_2025-04-19T19:42:11Z.tgz
command: restore db_backup_2025-04-19T19:42:11Z.tgz

But nothing helps. in the documentation everything is different everywhere. it is not clear how it should be. I also tried to request a docker-compose.yml from the grok but also unsuccessfully

[shaunakv1]

@deitch

want to add plus one to this. My restore to s3 compliant DigitalOcean Spaces isn't working either. Dump works just fine with the documented config patterns but restore doesn't.

Here's my backup and restore docker compose

  services:
    mysql-backup:
      image: databack/mysql-backup:master
      environment:
        DB_DEBUG: ${DB_DEBUG}
        DB_SERVER: ${DB_SERVER}
        DB_PORT: ${DB_PORT}
        DB_USER: ${DB_USER}
        DB_PASS: ${DB_PASS}
        DB_DUMP_INCLUDE: ${DB_DUMP_INCLUDE}
        SINGLE_DATABASE: ${SINGLE_DATABASE} 
        NO_DATABASE_NAME: ${NO_DATABASE_NAME}
        DB_DUMP_ONCE: ${DB_DUMP_ONCE}
        DB_DUMP_TRIGGERS_AND_FUNCTIONS: ${DB_DUMP_TRIGGERS_AND_FUNCTIONS}
        DB_DUMP_TARGET: ${DB_DUMP_TARGET}
        DB_DUMP_SAFECHARS: ${DB_DUMP_SAFECHARS}
        AWS_ACCESS_KEY_ID: ${AWS_ACCESS_KEY_ID}
        AWS_SECRET_ACCESS_KEY: ${AWS_SECRET_ACCESS_KEY}
        AWS_ENDPOINT_URL: ${AWS_ENDPOINT_URL}
        AWS_REGION: ${AWS_REGION}
      command: dump
    mysql-restore:
      image: databack/mysql-backup:master
      environment:
        DB_DEBUG: ${DB_DEBUG}
        RESTORE_DATABASE: ${RESTORE_DATABASE}
        DB_SERVER: ${DB_RESTORE_SERVER}
        DB_PORT: ${DB_RESTORE_PORT}
        DB_USER: ${DB_RESTORE_USER}
        DB_PASS: ${DB_RESTORE_PASS}
        DB_RESTORE_TARGET: ${DB_RESTORE_TARGET}
        AWS_ACCESS_KEY_ID: ${AWS_ACCESS_KEY_ID}
        AWS_SECRET_ACCESS_KEY: ${AWS_SECRET_ACCESS_KEY}
        AWS_ENDPOINT_URL: ${AWS_ENDPOINT_URL}
        AWS_REGION: ${AWS_REGION}
      command: ["restore", "db_backup_2025-04-26T04-25-56Z.tgz"]

relevant env variables

DB_DUMP_TARGET="s3://rlogx-db-backups/rlogx-prod/"
DB_RESTORE_TARGET="s3://rlogx-db-backups/rlogx-prod"
AWS_ENDPOINT_URL="https://nyc3.digitaloceanspaces.com"

Note rlogx-db-backups in the dump and restore is the bucket name that resolves with subdomain scheme. It works for dump but not for restore

Here's the full error log

dc up mysql-restore
[+] Running 1/1
 ✔ Container iac-mysql-restore-1  Created                                                                                                              0.0s
Attaching to mysql-restore-1
mysql-restore-1  | time="2025-04-29T03:15:39Z" level=debug msg="starting restore"
mysql-restore-1  | time="2025-04-29T03:15:39Z" level=info msg="beginning restore" run=a731f92f-6258-4203-ad0a-ed9b2a636ec6
mysql-restore-1  | time="2025-04-29T03:15:39Z" level=debug msg="restoring via s3 protocol, temporary file location /tmp/restorefile" run=a731f92f-6258-4203-ad0a-ed9b2a636ec6
mysql-restore-1  | Error: error restoring: failed to pull target &{{s3  %!s(*url.Userinfo=<nil>) rlogx-db-backups /rlogx-prod  %!s(bool=false) %!s(bool=false)   } %!s(bool=false)    }: failed to download file, operation error S3: GetObject, https response error StatusCode: 404, RequestID: tx00000b4758bb0507878d0-006810445b-2b885477-nyc3d, HostID: 2b885477-nyc3d-nyc3-zg04, NoSuchKey:
mysql-restore-1  | time="2025-04-29T03:15:39Z" level=fatal msg="error restoring: failed to pull target &{{s3  %!s(*url.Userinfo=<nil>) rlogx-db-backups /rlogx-prod  %!s(bool=false) %!s(bool=false)   } %!s(bool=false)    }: failed to download file, operation error S3: GetObject, https response error StatusCode: 404, RequestID: tx00000b4758bb0507878d0-006810445b-2b885477-nyc3d, HostID: 2b885477-nyc3d-nyc3-zg04, NoSuchKey: "
mysql-restore-1 exited with code 1

Something is wrong with restore and reading of config to generate the correct path.

[shaunakv1]

@deitch

I think I narrowed down the problem. In the Pull function in s3.go, there's an extra leading / that gets added to final path. This causes the restore file to be not found.

here's the logs with aws trace enabled

If I trim the leading / like so, the restore is able to pull the file from s3 bucket.

I can start a PR with that change but not sure if it will cause a regression somewhere else. What and why is that extra leading / being added? It's not in my config

[deitch]

I have a suspicion why. When you do:

	bucket, path := s.url.Hostname(), path.Join(s.url.Path, source)

the Path always starts with a /. So in the case of your example s3://rlogx-db-backups/rlogx-prod, it will give bucket = rlogx-db-backups and path = /rlogx-prod`. I believe (but am not positive), that when you call:

	n, err := downloader.Download(context.TODO(), f, &s3.GetObjectInput{
		Bucket: aws.String(bucket),
		Key:    aws.String(path),
	})

It assumes the Key is supposed to be /-free, and it prepends it. This should be provable by stepping through it with a debugger like dlv.

Aha, yes, look at the dump (Pull()) code here:

	// S3 always prepends a /, so if it already has one, it would become //
	// For some services, that is ok, but for others, it causes issues.
	key = strings.TrimPrefix(path.Join(key, target), "/")

So, yes, your change should be good. The slightly different language in Pull() is due to the difference in semantics between push and pull. Your change is correct.

Thanks for catching it, and please do open a PR.

[shaunakv1]

@deitch Thanks! here's the pr

#435

[shaunakv1]

@human211 I realize your question was slightly different but related. Once this PR is merged, the config you need to make this work is in my example above.

To be specific, in your docker compose file you want the dump file to be passed in the command like so:

command: ["restore", "db_backup_2025-04-26T04-25-56Z.tgz"]

Restore target should be the path to your dump file

DB_RESTORE_TARGET="s3://rlogx-db-backups/rlogx-prod"