From JP:
<--8<-->
At Datadog we have identified an issue for JFR recordings when the disk of the repository is full.

root@7269529ba912:/spring-petclinic/scripts# ls -l /tmp/2020_09_25_13_16_25_23/
total 4036
-rw-r--r-- 1 root root 3001528 Sep 25 13:17 2020_09_25_13_16_25.jfr
-rw-r--r-- 1 root root 1130496 Sep 25 13:19 2020_09_25_13_17_25.jfr
-rw-r--r-- 1 root root 0 Sep 25 13:19 2020_09_25_13_19_44.jfr
-rw-r--r-- 1 root root 0 Sep 25 13:20 2020_09_25_13_19_45.jfr
-rw-r--r-- 1 root root 0 Sep 25 13:21 2020_09_25_13_20_25.jfr
-rw-r--r-- 1 root root 0 Sep 25 13:21 2020_09_25_13_21_25.jfr

As you can see empty jfr files are created (metadata of the filesystem), but no data, expected. But from the API standpoint we have no information about the situation:
no error logged from Xlog:jfr for example
no exception throws from java Api when taking the snapshot for example.
The size of the Recording (jdk.jfr.Recording#getSize()) is also non zero
I assume that when a JFR file is written on disk, OS should report an error. This error should be reported one way or another.

Does that make sense?

To reproduce, it's relatively easy with a docker container that is filled with a command like:
dd if=/dev/zero of=largefile.dat bs=1G count=200

Thanks
Jean-Philippe
<--8<-->

It would be good to not only consider logging the problem, but also exposing it through the APIs, so that there is a chance for tools built on JFR to discover and act on the problem.