cas / gets — Compare-And-Swap

Code

# gets returns a unique cas token as the last value
gets counter
# VALUE counter 0 2 14
# 42
# END
#                 ^ cas_unique = 14

# cas — write only if cas_unique still matches
cas counter 0 0 2 14
43
# STORED       (token 14 still current)

# Another client modified counter between our gets and cas:
cas counter 0 0 2 14
44
# EXISTS       (token changed — someone else wrote first)

# Race-free increment pattern
gets counter        # 42, token=14
# ... compute new value ...
cas counter 0 0 2 14
43
# if EXISTS, retry: gets again, recompute, cas again

# Python (pymemcache)
from pymemcache.client.base import Client
c = Client(("localhost", 11211))
result = c.gets("counter")          # (value, cas)
ok = c.cas("counter", b"43", result[1], expire=0)
if not ok:
    pass  # retry

Explanation

cas (Compare-And-Swap) implements optimistic concurrency: gets fetches the value plus a cas_unique token; cas writes only if the token still matches, otherwise returns EXISTS. This avoids lost updates when two clients read-modify-write the same key. The standard retry pattern: gets → compute → cas → on EXISTS, loop. cas_unique is a server-side counter incremented on each write. Note cas does NOT bump exptime automatically — pass a new exptime if you want to extend ttl. For pure integer counters prefer incr/decr which are already atomic.

More Memcached Snippets