Hi Sergey! I'm using a Azure VM (ubuntu), mariadb ubuntu package 
Server version: 10.3.5-MariaDB-10.3.5+maria~xenial mariadb.org binary distribution


MariaDB [test]>  CREATE TABLE `table_without_compression` (
  `id` int(11) NOT NULL DEFAULT 0,
  `gmtime` decimal(22,7) NOT NULL DEFAULT 0.0000000,
  `data` blob NOT NULL,
  PRIMARY KEY (`id`,`gmtime`)
)  ENGINE=InnoDB DEFAULT CHARSET=latin1;
Query OK, 0 rows affected (0.385 sec)

MariaDB [test]>  CREATE TABLE `table_compressed` (
  `id` int(11) NOT NULL DEFAULT 0,
  `gmtime` decimal(22,7) NOT NULL DEFAULT 0.0000000,
  `data` blob COMPRESSED NOT NULL DEFAULT '',
  PRIMARY KEY (`id`,`gmtime`)
)  ENGINE=InnoDB DEFAULT CHARSET=latin1;
Query OK, 0 rows affected (0.385 sec)


MariaDB [test]> INSERT INTO table_without_compression VALUES (1,0,repeat('a',65535));
Query OK, 1 row affected (0.135 sec)

MariaDB [test]> INSERT INTO  table_without_compression VALUES (1,1,repeat('a',65536));
ERROR 1406 (22001): Data too long for column 'data' at row 1   <- JUST TO CHECK IF WE GOT MAX SIZE OF BLOB COLUMN, maybe that's the problem, internally compression consider the from/to column instead of a bigger column (long blob for example when using blob/longblob, i will test with TEXT/LONG TEXT and VARCHAR too)

MariaDB [test]> INSERT INTO table_compressed SELECT * FROM table_without_compression;
ERROR 1406 (22001): Data too long for column 'data' at row 1 <- that's the error

Should I include at Jira?