Feature request: migrate to java.time

[mostolog]

Hi

During some tests with Logstash 6.0.0-rc1 we have encountered a few errors parsing dates from different locales (some logs come in English, while other come in other languages) and, it looks like it could be related to Joda-Time being case-sensitive.

        ...
	mutate {
		add_field => {
			"date" => "11/Oct/2017:13:30:26 +0200"
			}
		}
	date {
		timezone => "Europe/Madrid"
		match => ["date", "dd/MMM/yyyy:HH:mm:ss Z"]
		locale => "es"
		tag_on_failure => "error_date"
	}
        ...

Specifically, If month pattern MMM is used and Joda expects 'oct' but found 'Oct'
Examples: https://discuss.elastic.co/t/timestamp-month-name-date-matching/25538

Is there any other function Logstash could use to be case-insensitive instead? https://stackoverflow.com/a/10797809

As stated by their own page (http://www.joda.org/joda-time/) migrating to java.time is advised.

As requested in #7149, I would like to ask for a library parameter to enable the usage of java.time instead of Joda-Time in the event of backwards-compatibility issues. Otherwise a complete migration would be gracefully welcomed.

On the other hand, it would be really good knowing if there's a workaround, other than lowercasing month strings in order Joda being able to parse date in a case-insensitive manner.

Thanks,
Regards

CC: @magnusbaeck

[jordansissel]

This feature request is about an implementation detail, not a behavior. Let's focus on behaviors.

You mention that "Oct" fails to parse, but "oct" is successful, and you say java.time will solve this.

However, I find that it does not solve it.

Reproducing your case, using MMM with locale "es", I find both Joda-Time and java.time are reject "Oct" and accept "oct" when locale is "es"

• Joda-Time "oct" vs "Oct" with locale "es"

# "oct" - works.
>> org.joda.time.format.DateTimeFormat.forPattern("MMM").withLocale(java.util.Locale.forLanguageTag("es")).parseDateTime("oct").getMonthOfYear
=> 10

# "Oct" fails
>> org.joda.time.format.DateTimeFormat.forPattern("MMM").withLocale(java.util.Locale.forLanguageTag("es")).parseDateTime("Oct").getMonthOfYear
Java::JavaLang::IllegalArgumentException: Invalid format: "Oct"

• java.time

# "oct" works
>> java.time.format.DateTimeFormatter.ofPattern("MMM").withLocale(java.util.Locale.forLanguageTag("es")).parse("oct").get(java.time.temporal.ChronoField::MONTH_OF_YEAR)
=> 10

# "Oct" fails
>> java.time.format.DateTimeFormatter.ofPattern("MMM").withLocale(java.util.Locale.forLanguageTag("es")).parse("Oct").get(java.time.temporal.ChronoField::MONTH_OF_YEAR)
Java::JavaTimeFormat::DateTimeParseException: Text 'Oct' could not be parsed at index 0

[mostolog]

Hi @jordansissel

You are right, this was about an implementation detail at the beginning

As months (and days) in spanish have first letter capital, parsing uppercased strings should work.
eg: https://stackoverflow.com/a/38251032

So, this was an issue to fix how current implementation (which uses joda.time) does the parsing.

However. as joda.time is deprecated, I then suggested ensuring this case-sensitive issue would be fixed when/if migrated to java.time.