Re: Having a hard time understanding time zone

From: Robert DiFalco <robert(dot)difalco(at)gmail(dot)com>
To: Dave Cramer <pg(at)fastcrypt(dot)com>
Cc: List <pgsql-jdbc(at)postgresql(dot)org>
Subject: Re: Having a hard time understanding time zone
Date: 2014-04-09 00:31:16
Message-ID: CAAXGW-y==FOGGB7jVWhDsaugXV20qeXUcgZSiteQ7jfMXOKi-A@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-jdbc

Okay, I've got it. I went through the Driver code and apparently in the 9.3
ConnectionFactoryImpl.java the driver does send the JVM timezone as part of
the start-up packets. Seems like this is what is causing server side
function executions (like NOW()) to use the JVM time zone, even if they
aren't executed explicitly by the driver itself. This looks to be
equivalent to "SET TIME ZONE <jvm_timezone>" for each session.

This happens around line 167 for my version of the driver.

On Tue, Apr 8, 2014 at 5:10 PM, Robert DiFalco <robert(dot)difalco(at)gmail(dot)com>wrote:

> If you look at the test case you will see that I get the TIMESTAMP with
> PSQL. So I can only assume from the above code that the server does now
> somehow to execute NOW() in the client's timezone for TIMESTAMP WITHOUT
> TIMEZONE fields. You can try it yourself. Somehow even though
> CURRENT_TIMESTAMP is executing on the server, it is picking up the JVM
> timezone. Since TIMESTAMPTZ is UTC this does not have the same issues.
>
>
> On Mon, Apr 7, 2014 at 11:51 AM, Dave Cramer <pg(at)fastcrypt(dot)com> wrote:
>
>> Robert,
>>
>> As I said neither Java nor JDBC do justice to date/time etal.
>>
>> 1) You need to know that the JDBC spec does not have two timestamp types
>> so we do the best we can.
>>
>> Without digging into the code I'm betting that if the timestamp comes
>> back with a timezone we use it. if it doesn't we use the JVM's
>>
>>
>>
>> Dave Cramer
>>
>> dave.cramer(at)credativ(dot)ca
>> http://www.credativ.ca
>>
>>
>> On 7 April 2014 14:43, Robert DiFalco <robert(dot)difalco(at)gmail(dot)com> wrote:
>>
>>> Okay, I've got this narrowed down and it seems crazy to me.
>>>
>>>
>>> => show time zone;
>>> TimeZone
>>> ----------
>>> UTC
>>> (1 row)
>>>
>>> => CREATE TABLE test (id INTEGER, ts TIMESTAMP DEFAULT
>>> CURRENT_TIMESTAMP, tswz TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP);
>>> CREAT TABLE
>>>
>>> => insert into test(id) values(1);
>>> INSERT 0 1
>>> warm-headland-9732::OLIVE=> select * from test;
>>> id | ts | tswz
>>> ----+----------------------------+-------------------------------
>>> 1 | 2014-04-07 18:29:30.990622 | 2014-04-07 18:29:30.990622+00
>>> (1 row)
>>>
>>>
>>>
>>>
>>> This is all as I would expect it to be. But if I run this same insert
>>> from my JVM I get this!
>>>
>>> => select * from test;
>>> id | ts | tswz
>>> ----+----------------------------+-------------------------------
>>> 1 | 2014-04-07 18:29:30.990622 | 2014-04-07 18:29:30.990622+00
>>> 1 | 2014-04-07 11:32:28.692483 | 2014-04-07 18:32:28.692483+00
>>> (2 rows)
>>>
>>>
>>> Note that the ts defaults to a PDT value while the tswz defaults to the
>>> proper UTC value. So I'm stumped. What is special about Java that causes
>>> CURRENT_TIMESTAMP to evaluate differently for a TIMESTAMP WITHOUT TIME ZONE
>>> field? I'm not even specifying the "ts" field from Java, I am just letting
>>> it default? Any ideas? The same query from PSQL gives different results
>>> than from my JVM for default values. Here is the Java code so you can
>>> recreate the test.
>>>
>>> @Test
>>> public void testTimestamp() throws Exception {
>>>
>>> try ( Connection con = dataSource.getConnection() ) {
>>> try ( Statement st = con.createStatement() ) {
>>> st.execute( "insert into test(id) values(1)");
>>> }
>>>
>>> con.commit();
>>> }
>>> }
>>>
>>>
>>> Note that the Java insert does not specify "ts" or "tswz" so there is no
>>> time transformations going on.
>>>
>>> R.
>>>
>>>
>>>
>>> On Sun, Apr 6, 2014 at 1:45 PM, Dave Cramer <pg(at)fastcrypt(dot)com> wrote:
>>>
>>>> Robert,
>>>>
>>>> The driver certainly doesn't do anything to a statement like
>>>>
>>>> insert into foo (datecol) values (now())
>>>>
>>>> it does use the calendar of the vm to get the string representation.
>>>>
>>>> Time, and date are two things not very well handled in java or JDBC.
>>>> What is the exact column type and what does it store in the database
>>>>
>>>> But to answer your question you can do:
>>>>
>>>> stmt.setTimestamp(1, t, Calendar.getInstance( TimeZone.getTimeZone("UTC")))
>>>>
>>>>
>>>> Dave Cramer
>>>>
>>>> dave.cramer(at)credativ(dot)ca
>>>> http://www.credativ.ca
>>>>
>>>>
>>>> On 6 April 2014 13:56, Robert DiFalco <robert(dot)difalco(at)gmail(dot)com> wrote:
>>>>
>>>>> Does the JDBC driver set the timezone to the origin timezone for each
>>>>> statement?
>>>>>
>>>>> I have a date column in a table. My Postgres server is running in UTC.
>>>>> My java app is running in "America/Los_Angeles".
>>>>>
>>>>> I would expect a DEFAULT column of NOW() to insert the current UTC
>>>>> time. While if I specify the time with new Date() from Java I would expect
>>>>> the Java timezone.
>>>>>
>>>>> But oddly both set the date field to the localized time in Java and
>>>>> not the UTC time.
>>>>>
>>>>> This makes me think that the driver is somehow forcing the session
>>>>> timezone.
>>>>>
>>>>> If so is there any way to make the driver communicate with the server
>>>>> in UTC?
>>>>>
>>>>> Thanks!
>>>>>
>>>>
>>>>
>>>
>>
>

In response to

Responses

Browse pgsql-jdbc by date

  From Date Subject
Next Message Dave Cramer 2014-04-09 10:19:23 Re: Having a hard time understanding time zone
Previous Message Robert DiFalco 2014-04-09 00:10:42 Re: Having a hard time understanding time zone