As Andrey noted, this report is the cmin half of the issue whose
cmax half was fixed in 8c67d29fd51. Confirmed on current master
(36f7330b8b2) with a slightly different table shape, and there is
an existing patch on pgsql-hackers [1], which fixes the crash in
my testing. I have posted reproduction results, a WAL-level
confirmation of the diagnosis, and review comments in that
thread [2].
[1] https://www.postgresql.org/message-id/CAHgHdKu5e3XY5e90Tuaxq_R4WrKxSV734Q%2BLwo5y39Omp2A-Gg%40mail.gmail.com
[2] https://www.postgresql.org/message-id/ME0P300MB05682FD742273EC64362A8A1C6D42%40ME0P300MB0568.AUSP300.PROD.OUTLOOK.COM
Thanks,
Bingshuai Li