Linux – Server timezone changed to PHT but logs still written in MST

Transaction logs contain the binary deltas of modifications applied to a database since the last log truncation; the keyword here is "binary": they don't contain SQL queries or something similar, but are actually more akin to binary patches to be applied to a program.

For this reason, they can be replayed only on the exact physical database files they were originally linked to; replaying them on another database (even with the same schema) would be exactly like applying a patch to a different executable to the one it was written for: not possible at all.

You also can't replay them on the same database, if that gets modified; i.e., you can't restore the database from an old backup, bring it online, make any modification at all to it and then replay the logs against it; you actually lose any log-replaying capability as soon as you bring the database fully online (there's even a specific flag in SQL Server restore operations for that).

Does your application check the time by checking Unix time (time_t, seconds since the epoch), or does it check just human-expressed time, neglecting the timezone?

If the latter, I fear you may have made a coding error. The timezone can be set and changed on a process-by-process basis, as easily as changing the TZ environment variable; try

export TZ=BST
date
export TZ=EST
date

Note that this is an environment variable, and thus inheritable by the children of the running process. If the java process happens to be restarted by someone who has TZ set to something you didn't expect, and your startup file doesn't force TZ, then java will be running with an unexpected timezone.

TZ is by no means the only way of setting the timezone; I wanted to mention this to show you that Unix is designed to allow the timezone to be a personal convenience, much like your shell prompt or your PATH. If you need a process to have a particular view of timezone, you need to make sure that it sets that when it starts up.