Why do gaps/skips occur in GTID sets? (no replies)

MySQL server 5.6.35-log (community)

I have two master servers replicating to one another. They are identical in every way including their configuration. I keep getting gaps in the GTID sets and I don't know why. This presents a real problems when off-line SLAVES re-attach. They complain that sequences have been purged when in fact they haven't been purged - they just don't exist.

Here is a real-world example:

mysql01> SHOW SLAVE STATUS;

EXECUTED_GTUD_SET
278bfda1-b93d-11e4-801b-14feb5d284bc:1-1429638,
819c985c-d384-11e3-a621-00259002979a:1-159919

mysql02> SHOW SLAVE STATUS;

EXECUTED_GTUD_SET
278bfda1-b93d-11e4-801b-14feb5d284bc:1-1429638,
819c985c-d384-11e3-a621-00259002979a:1-154865:154874-154880:154882-154883:154885:154887:154889:154891:154893-159919

Now I have to inject empty transaction so the slaves will connect and replicate.

Why does this occur?