Core Java

What’s Even Harder Than Dates and Timezones? Dates and Timezones in SQL / JDBC!

There was an interesting discussion recently on the jOOQ mailing list about jOOQ’s current lack of out-of-the-box support for TIMESTAMP WITH TIME ZONE data types.

No one said that date, time and timezones are easy! There’s an amusing piece here, which I recommend reading: Falsehoods programmers believe about time

And when that’s not enough, read also: More falsehoods programmers believe about time

I personally like the bit about programmers erroneously believing that “Unix time is the number of seconds since Jan 1st 1970.”… unix time doesn’t have a way to represent leap seconds ;)

Back to JDBC

Here’s an interesting Stack Overflow answer by Mark Rotteveel, the Jaybird developer (Firebird JDBC driver): Is java.sql.Timestamp timezone specific?

Mark’s explanation can be observed as follows (I’m using PostgreSQL here):

Connection c = getConnection();
Calendar utc = Calendar.getInstance(
    TimeZone.getTimeZone("UTC"));

try (PreparedStatement ps = c.prepareStatement(
    "select"
  + "  ?::timestamp,"
  + "  ?::timestamp,"
  + "  ?::timestamp with time zone,"
  + "  ?::timestamp with time zone"
)) {

    ps.setTimestamp(1, new Timestamp(0));
    ps.setTimestamp(2, new Timestamp(0), utc);
    ps.setTimestamp(3, new Timestamp(0));
    ps.setTimestamp(4, new Timestamp(0), utc);

    try (ResultSet rs = ps.executeQuery()) {
        rs.next();

        System.out.println(rs.getTimestamp(1) 
                 + " / " + rs.getTimestamp(1).getTime());
        System.out.println(rs.getTimestamp(2, utc)
                 + " / " + rs.getTimestamp(2, utc).getTime());
        System.out.println(rs.getTimestamp(3) 
                 + " / " + rs.getTimestamp(3).getTime());
        System.out.println(rs.getTimestamp(4, utc)
                 + " / " + rs.getTimestamp(4, utc).getTime());
    }
}

The above program uses all permutations of using timezones and not using timezones in Java and in the DB, and the output is always the same:

1970-01-01 01:00:00.0 / 0
1970-01-01 01:00:00.0 / 0
1970-01-01 01:00:00.0 / 0
1970-01-01 01:00:00.0 / 0

As you can see, in each case, the UTC timestamp 0 was correctly stored and retrieved from the database. My own locale is Switzerland, thus CET / CEST, which was UTC+1 at Epoch, which is what is getting output on Timestamp.toString().

Things get interesting when you use timestamp literals, both in SQL and/or in Java. If you replace the bind variables as such:

Timestamp almostEpoch = Timestamp.valueOf("1970-01-01 00:00:00");

ps.setTimestamp(1, almostEpoch);
ps.setTimestamp(2, almostEpoch, utc);
ps.setTimestamp(3, almostEpoch);
ps.setTimestamp(4, almostEpoch, utc);

This is what I’m getting on my machine, again in CET / CEST

1970-01-01 00:00:00.0 / -3600000
1970-01-01 00:00:00.0 / -3600000
1970-01-01 00:00:00.0 / -3600000
1970-01-01 00:00:00.0 / -3600000

I.e. not Epoch, but the timestamp literal that I sent to the server in the first place. Observe that the four combinations of binding / fetching still always produce the same timestamp.

Let’s see what happens if the session writing to the database uses a different timezone (let’s assume you’re in PST) than the session fetching from the database (I’m using again CET or UTC). I’m running this program:

Calendar utc = Calendar.getInstance(
    TimeZone.getTimeZone("UTC"));

Calendar pst = Calendar.getInstance(
    TimeZone.getTimeZone("PST"));

try (PreparedStatement ps = c.prepareStatement(
    "select"
  + "  ?::timestamp,"
  + "  ?::timestamp,"
  + "  ?::timestamp with time zone,"
  + "  ?::timestamp with time zone"
)) {

    ps.setTimestamp(1, new Timestamp(0), pst);
    ps.setTimestamp(2, new Timestamp(0), pst);
    ps.setTimestamp(3, new Timestamp(0), pst);
    ps.setTimestamp(4, new Timestamp(0), pst);

    try (ResultSet rs = ps.executeQuery()) {
        rs.next();

        System.out.println(rs.getTimestamp(1)
                 + " / " + rs.getTimestamp(1).getTime());
        System.out.println(rs.getTimestamp(2, utc)
                 + " / " + rs.getTimestamp(2, utc).getTime());
        System.out.println(rs.getTimestamp(3)
                 + " / " + rs.getTimestamp(3).getTime());
        System.out.println(rs.getTimestamp(4, utc)
                 + " / " + rs.getTimestamp(4, utc).getTime());
    }
}

It yields this output:

1969-12-31 16:00:00.0 / -32400000
1969-12-31 17:00:00.0 / -28800000
1970-01-01 01:00:00.0 / 0
1970-01-01 01:00:00.0 / 0

The first timestamp was Epoch stored as PST (16:00), then the timezone information was removed by the database, which turned Epoch into the local time you had at Epoch (-28800 seconds / -8h), and that’s the information that is really stored.

Now, when I’m fetching this time from my own timezone CET, I will still want to get the local time (16:00). But in my timezone, this is no longer -28800 seconds, but -32400 seconds (-9h). Quirky enough?

Things go the other way round when I’m fetching the stored local time (16:00), but I’m forcing the fetch to happen in UTC, which will produce the timestamp that you’ve stored, originally in PST (-28800 seconds). But when printing this timestamp (-28800 seconds) in my timezone CET, this will now be 17:00.

When we use the TIMESTAMP WITH TIME ZONE data type in the database, the timezone is maintained (PST), and when I fetch the Timestamp value, no matter if using CET or UTC, I will still get Epoch, which was safely stored to the database, printed out as 01:00 in CET.

Whew.

TL;DR:

When using jOOQ’s, if the correct UTC timestamp matters to you, use TIMESTAMP WITH TIMEZONE, but you’ll have to implement your own data type Binding, because jOOQ currently doesn’t support that data type. Once you use your own data type Binding, you can also use Java 8’s time API, which better represents these different types than java.sql.Timestamp + the ugly Calendar.

If the local time matters to you, or if you’re not operating across time zones, you’re fine using TIMESTAMP and jOOQ’s Field<Timestamp>.

Lucky you, if you’re like me, operating in a very small country with a single time zone where most local software just doesn’t run into this issue.

Lukas Eder

Lukas is a Java and SQL enthusiast developer. He created the Data Geekery GmbH. He is the creator of jOOQ, a comprehensive SQL library for Java, and he is blogging mostly about these three topics: Java, SQL and jOOQ.
Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

2 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Ruurd
Ruurd
8 years ago

UNIX time does not have support for leap seconds? Why should it? It is seconds since epoch. If you use that to convert it to a struct tm then yes that takes leap seconds into account. Look at the max value of seconds in struct tm. So where does the falsehood that UNIX time does not support leap seconds?

Lukas Eder
8 years ago
Reply to  Ruurd

There is no way to represent a time like 23:59:60 in a unix timestamp. It has no representation. It doesn’t exist. So, in other words, unix timestamps is NOT seconds since epoch, because the leap seconds are skipped. See #15 of http://infiniteundo.com/post/25509354022/more-falsehoods-programmers-believe-about-time “Unix time is the number of seconds since Jan 1st 1970.”

Back to top button