Thread: Password Encryption and Connection Issues
Hello all
We have recently upgraded our PostgreSQL instances from version 13 to 16. During the upgrade, we also changed the password_encryption setting in postgresql.conf to scram-sha-256.Before the upgrade, we used pg_dumpall --roles-only to export all users and their MD5-hashed passwords. After the upgrade, we executed this SQL script to restore the users, and all users with their MD5 hashes were recreated successfully.
However, we observed that:
New users created under the
scram-sha-256encryption setting have passwords starting withSCRAM-SHA-256$4096:inpg_authid.The imported users still have passwords in the MD5 format, e.g.,
md5a33e074800fe59f4ec8a123d0085d0e9.Our
pg_hba.confstill usesmd5as the authentication method.
As a result, some users are able to connect, while others cannot.
My questions are:
Is it expected behavior that users created with
scram-sha-256passwords can still connect viamd5inpg_hba.conf?Under the current settings, is it still possible to use MD5-style password hashes for user creation? How does PostgreSQL treat this compatibility?
- In such a case, what would be the recommended approach or best practice to follow during upgrades in order to avoid this kind of issue?
Thank you in advance for your support.
Best regards,
Alpaslan
On Wed, Jul 9, 2025 at 6:57 AM Alpaslan AKDAĞ <alpaslanakdag@gmail.com> wrote:
- In such a case, what would be the recommended approach or best practice to follow during upgrades in order to avoid this kind of issue?
This is all described quite clearly in the documentation, including the upgrade procedure in the final paragraph.
Given that pg_hba.conf still uses md5 I'm a bit confused regarding the claim of some people being unable to authenticate; but you've provided insufficient data to diagnose. In any case, hopefully you can just change all passwords to use scram and move on.
David J.
On Wed, Jul 9, 2025 at 9:57 AM Alpaslan AKDAĞ <alpaslanakdag@gmail.com> wrote:
Is it expected behavior that users created withscram-sha-256passwords can still connect viamd5inpg_hba.conf?
Yes. From the docs:
To ease transition from themd5method to the newer SCRAM method, ifmd5is specified as a method inpg_hba.confbut the user's password on the server is encrypted for SCRAM (see below), then SCRAM-based authentication will automatically be chosen instead.
You can think of "md5" inside pg_hba.conf as "md5 or better"
As a result, some users are able to connect, while others cannot.
Can you expand on this? Nothing you have done should be preventing logins, as far as I can tell.
Best solution: Upgrade everyone to scram, then change md5 to scram in pg_hba.conf and never look back.
Cheers,
Greg
--
Crunchy Data - https://www.crunchydata.com
Enterprise Postgres Software Products & Tech Support
On Wed, Jul 9, 2025 at 10:59 AM Greg Sabino Mullane <htamfids@gmail.com> wrote:
On Wed, Jul 9, 2025 at 9:57 AM Alpaslan AKDAĞ <alpaslanakdag@gmail.com> wrote:Is it expected behavior that users created withscram-sha-256passwords can still connect viamd5inpg_hba.conf?Yes. From the docs:To ease transition from themd5method to the newer SCRAM method, ifmd5is specified as a method inpg_hba.confbut the user's password on the server is encrypted for SCRAM (see below), then SCRAM-based authentication will automatically be chosen instead.You can think of "md5" inside pg_hba.conf as "md5 or better"As a result, some users are able to connect, while others cannot.Can you expand on this? Nothing you have done should be preventing logins, as far as I can tell.Best solution: Upgrade everyone to scram, then change md5 to scram in pg_hba.conf and never look back.
That requires setting the password to null and then recreating the password, no? Otherwise IIRC, changing an md5 password leaves the new password also in md5 format.
Death to <Redacted>, and butter sauce.
Don't boil me, I'm still alive.
<Redacted> lobster!
On 7/9/25 06:56, Alpaslan AKDAĞ wrote: > Hello all > > As a result, some users are able to connect, while others cannot. What client is being used and what version of said client? > Best regards, > > Alpaslan > > -- Adrian Klaver adrian.klaver@aklaver.com
On Wed, Jul 9, 2025 at 11:11 AM Adrian Klaver <adrian.klaver@aklaver.com> wrote:
On 7/9/25 06:56, Alpaslan AKDAĞ wrote:
> Hello all
>
> As a result, some users are able to connect, while others cannot.
What client is being used and what version of said client?
This is a salient point:clients from the pre-PG10 can only connect using md5. Thus, we have to use md5 hashes even in PG16. 😭
Death to <Redacted>, and butter sauce.
Don't boil me, I'm still alive.
<Redacted> lobster!
On Wed, Jul 9, 2025 at 8:09 AM Ron Johnson <ronljohnsonjr@gmail.com> wrote:
That requires setting the password to null and then recreating the password, no?
You might want to verify that claim, and suggest a doc patch or bug fix if you find it to be true - I sure don't see anything that remotely suggests this.
David J.
On Wed, 2025-07-09 at 11:09 -0400, Ron Johnson wrote: > > Best solution: Upgrade everyone to scram, then change md5 to scram > > in pg_hba.conf and never look back. > > That requires setting the password to null and then recreating the > password, no? Otherwise IIRC, changing an md5 password leaves the > new password also in md5 format. No. The hashing algorithm chosen depends only on the current setting of "password_encryption", not on the hashing algorithm chosen for the previous password. Yours, Laurenz Albe
On Wed, Jul 9, 2025 at 11:26 AM David G. Johnston <david.g.johnston@gmail.com> wrote:
On Wed, Jul 9, 2025 at 8:09 AM Ron Johnson <ronljohnsonjr@gmail.com> wrote:That requires setting the password to null and then recreating the password, no?You might want to verify that claim, and suggest a doc patch or bug fix if you find it to be true - I sure don't see anything that remotely suggests this.
You're right: I didn't recall properly.
Death to <Redacted>, and butter sauce.
Don't boil me, I'm still alive.
<Redacted> lobster!
Best solution: Upgrade everyone to scram, then change md5 to scram in pg_hba.conf and never look back.
To expand more on the "upgrade everyone to scram", that means force all users to set a new password while using scram (which should be the default). You can do it yourself by getting a list of users and changing their passwords inside psql:
-- List all users still stuck in md5-land:
greg=# select rolname from pg_authid where rolpassword ~ '^md5'
alice
eve
mallory
(3 rows)
-- Just in case, force use of scram
greg=# set password_encryption = 'scram-sha-256';
SET
-- Reset each user's password to some strong password of your choice:
greg=# \password alice
Enter new password for user "alice":
Enter it again:
Enter it again:
-- Repeat the above until this query returns no rows:
select rolname from pg_authid where rolpassword ~ '^md5'
Cheers,
Greg