FULL PRODUCT VERSION :
Java(TM) SE Runtime Environment 1.7.0_80-b15

ADDITIONAL OS VERSION INFORMATION :
Linux 3.19.0-75-generic #83-Ubuntu SMP Wed Nov 9 23:37:58 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

A DESCRIPTION OF THE PROBLEM :
The following code:

Calendar cal = new GregorianCalendar(2000, 0, 1);
long testCalOne = cal.getTimeInMillis();
cal.setTimeZone(TimeZone.getTimeZone("UTC"));
long testCalTwo = cal.getTimeInMillis();

Calendar cal2 = new GregorianCalendar(2000, 0, 1);
cal2.setTimeZone(TimeZone.getTimeZone("UTC"));
long testCalThree = cal2.getTimeInMillis();

System.out.println(testCalOne + ", " + testCalTwo + ", " + testCalThree);

Prints 946681200000, 946681200000, 946684800000
What is expected is 946684800000, 946684800000, 946684800000
This is done on a machine on a TimeZone GMT +1.

What this means is that getTimeInMillis() is bugged and doesn't return the amount of milliseconds since 1970-01-01 in UTC, but instead in the Calendars timezone. The documentation references it returning it in UTC.

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Set timezone to a timezone different from UTC.

Run the following code

Calendar cal = new GregorianCalendar(2000, 0, 1);
long testCalOne = cal.getTimeInMillis();
cal.setTimeZone(TimeZone.getTimeZone("UTC"));
long testCalTwo = cal.getTimeInMillis();

Calendar cal2 = new GregorianCalendar(2000, 0, 1);
cal2.setTimeZone(TimeZone.getTimeZone("UTC"));
long testCalThree = cal2.getTimeInMillis();

System.out.println(testCalOne + ", " + testCalTwo + ", " + testCalThree);

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
946684800000, 946684800000, 946684800000

On the System.out
ACTUAL -
946681200000, 946681200000, 946684800000

On the System.out. This is in a timezone +1 hour relative to UTC.

REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
Calendar cal = new GregorianCalendar(2000, 0, 1);
long testCalOne = cal.getTimeInMillis();
cal.setTimeZone(TimeZone.getTimeZone("UTC"));
long testCalTwo = cal.getTimeInMillis();

Calendar cal2 = new GregorianCalendar(2000, 0, 1);
cal2.setTimeZone(TimeZone.getTimeZone("UTC"));
long testCalThree = cal2.getTimeInMillis();

System.out.println(testCalOne + ", " + testCalTwo + ", " + testCalThree);

Assert.assertEquals(946684800000L, testCalOne);
Assert.assertEquals(946684800000L, testCalTwo);
Assert.assertEquals(946684800000L, testCalThree);
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
Replace
cal.getTimeInMillis()

with

cal.clear(Calendar.ZONE_OFFSET);
cal.setTimeZone(TimeZone.getTimeZone("UTC"));
cal.getTimeInMillis()

Keep in mind this will change cal.
The clear call is necessary because setTimeZone shouldn't affect the getTimeInMillis method, but does due to this bug.