Logstash/ElasticSearch: guesses wrong for datatype for field - elasticsearch

Question

The log files I'm trying to import into Logstash contain a field that sometimes looks like a date/time and sometimes does not. Unfortunately, the first occurrence looked like a date/time and someone (logstash or elasticsearch) decided to define the field as a date/time. When trying to import a later log record, Elasticsearch has an exception:
Failed to execute [index ...]
org.elasticsearch.index.mapper.MapperParsingException: Failed to parse [#fields.field99]
at org.elasticsearch.index.mapper.core.AbstractFieldMapper.parse(AbstractFieldMapper.java:320)
at org.elasticsearch.index.mapper.object.ObjectMapper.serializeValue(ObjectMapper.java:587)
...
Caused by: java.lang.IllegalArgumentException: Invalid format: "empty"
at org.elasticsearch.common.joda.time.format.DateTimeFormatter.parseMillis(DateTimeFormatter.java:747)
...
Question: How do I tell logstash/elasticsearch to not define this field as a date/time? I would like all the fields from my log (except the one explicit timestamp field) to be defined as just text.
Question: it appears that logstash gives up trying to import records from the log file after seeing this one that elasticsearch throws an exception on. How can I tell logstash to ignore this exception and keep trying to import the other records from the log file?

Answer 1

I found the answer to my first question myself.
Before adding data through Logstash, I had to set the defaults for Elasticsearch to treat the field as "string" instead of "date".
I did this by creating a defaults.js file like this:
{
"template": "logstash-*",
"mappings": {
`"_default_"`: {
"dynamic_templates": [{
"fields_template": {
"mapping": { "type": "string" },
"path_match": "#fields.*"
}
}]
}
}
}
and telling Elasticsearch to use it before adding any data through Logstash:
curl -XPUT 'http://localhost:9200/_template/template_logstash/' -d #defaults_for_elasticsearch.js
Hope this helps someone else.