BUG #15188: wrong count "tuples removed" in autovacuum log file

The following bug has been logged on the website:

Bug reference: 15188
Logged by: Anton G.
Email address: a(dot)glushakov86(at)gmail(dot)com
PostgreSQL version: 9.6.5
Operating system: Linux RedHat
Description:

Hi.
I have table with high rate update (~2500 updates/sec)
autovacumm processing this table every ~10 min, and log work result.
But in log i see that was vacuumed about 40-45k tuples, but in
pg_stat_user_tables i see that was removed more than 1,5 million dead tuples
(when autovacuum start in table about 1 mln dead tuples)
Here is two result from pg_stat_user_tables - before autovacuum start and
after , and autovacuum log:

# select now(),* from pg_stat_user_tables where relname = 't';
-[ RECORD 1 ]-------+------------------------------
now | 2018-04-26 13:36:07.707784+03
relid | 32170
schemaname | public
relname | t
seq_scan | 151371
seq_tup_read | 132672944
idx_scan | 4525649653
idx_tup_fetch | 9402703531
n_tup_ins | 33785210
n_tup_upd | 4502119094
n_tup_del | 0
n_tup_hot_upd | 0
n_live_tup | 33815482
n_dead_tup | 1926735
n_mod_since_analyze | 492583
last_vacuum | 2018-04-25 07:51:33.307364+03
last_autovacuum | 2018-04-26 13:21:53.931788+03
last_analyze | 2018-04-01 19:40:08.063981+03
last_autoanalyze | 2018-04-26 13:32:15.845189+03
vacuum_count | 5
autovacuum_count | 8628
analyze_count | 2
autoanalyze_count | 13407

# select now(),* from pg_stat_user_tables where relname = 't';
-[ RECORD 1 ]-------+------------------------------
now | 2018-04-26 13:38:07.320179+03
relid | 32170
schemaname | public
relname | t
seq_scan | 151371
seq_tup_read | 132672944
idx_scan | 4525898773
idx_tup_fetch | 9403224417
n_tup_ins | 33789071
n_tup_upd | 4502366855
n_tup_del | 0
n_tup_hot_upd | 0
n_live_tup | 33830290
n_dead_tup | 296035
n_mod_since_analyze | 49793
last_vacuum | 2018-04-25 07:51:33.307364+03
last_autovacuum | 2018-04-26 13:37:28.072735+03
last_analyze | 2018-04-01 19:40:08.063981+03
last_autoanalyze | 2018-04-26 13:37:44.353704+03
vacuum_count | 5
autovacuum_count | 8629
analyze_count | 2
autoanalyze_count | 13408

2018-04-26 13:37:28 MSK 12076 LOG: automatic vacuum of table "t": index
scans: 1
pages: 0 removed, 549695 remain, 0 skipped due to pins, 43363
skipped frozen
tuples: 53866 removed, 33836682 remain, 13483 are dead but not yet
removable
buffer usage: 801664 hits, 1280506 misses, 7528 dirtied
avg read rate: 96.220 MB/s, avg write rate: 0.566 MB/s
system usage: CPU 10.54s/16.85u sec elapsed 103.96 sec

Autovacuum log display that only 53866 dead tuples was removed (+13483
cannot), but in real was delete > 1,5 million
I think this is some problem with logging algorithm
Thank