flowcontrol value somehow corrupted in memcached

Observed that using different flow control values between tests eventually results in incorrect max_buffer size

From Variable: const uint8_t *key (memcached.c:4688)

connection_buffer_size = 64
$1 = (const uint8_t *) 0x7fd7c7d17c18 "connection_buffer_size64"

connection_buffer_size= 29
$6 = (const uint8_t *) 0x7fb66f36b018 "connection_buffer_size29"

then back to...

connection_buffer_size = 64
$16 = (const uint8_t *) 0x7fb66f387218 "connection_buffer_size647c6e5d6d8c561204a7dcd"

• results in 647 and I'm getting more data than wanted into buffer.

I did a tcpdump to check I was sending right thing and looks ok,

127.0.0.1.64731 > 127.0.0.1.12000: Flags [P.], cksum 0xfe58 (incorrect -> 0xb3d4), seq 127447519:127447567, ack 1903305022, win 9185, options [nop,nop,TS val 230838420 ecr 230838420], length 48
0x0000: 4500 0064 82ca 4000 4006 0000 7f00 0001 E..d..@.@.......
0x0010: 7f00 0001 fcdb 2ee0 0798 b1df 7172 213e ............qr!>
0x0020: 8018 23e1 fe58 0000 0101 080a 0dc2 5094 ..#..X........P.
0x0030: 0dc2 5094 805e 0016 0000 0000 0000 0018 ..P..^..........
0x0040: ff00 0005 0000 0000 0000 0000 636f 6e6e ............conn
0x0050: 6563 7469 6f6e 5f62 7566 6665 725f 7369 ection_buffer_si
0x0060: 7a65 3634 ze64