bug: config_util.lua:99: attempt to index local 'item' (a boolean value)

[Sn0rt]

Current Behavior

2023/07/11 10:25:26 [error] 10644#10644: *95142022 [lua] config_etcd.lua:659: failed to fetch data from etcd: /usr/local/apisix/apisix/core/config_util.lua:99: attempt to index local 'item' (a boolean value)
stack traceback:
        /usr/local/apisix/apisix/core/config_util.lua:99: in function 'fire_all_clean_handlers'
        /usr/local/apisix/apisix/core/config_etcd.lua:376: in function 'sync_data'
        /usr/local/apisix/apisix/core/config_etcd.lua:606: in function </usr/local/apisix/apisix/core/config_etcd.lua:596>
        [C]: in function 'xpcall'
        /usr/local/apisix/apisix/core/config_etcd.lua:596: in function </usr/local/apisix/apisix/core/config_etcd.lua:574>,  etcd key: /apisix/routes, context: ngx.timer

Expected Behavior

no error log

Error Logs

No response

Steps to Reproduce

APISIX: 3.2.0
ETCD: 3.5

Environment

• APISIX version (run apisix version):
• Operating system (run uname -a):
• OpenResty / Nginx version (run openresty -V or nginx -V):
• etcd version, if relevant (run curl http://127.0.0.1:9090/v1/server_info):
• APISIX Dashboard version, if relevant:
• Plugin runner version, for issues related to plugin runners:
• LuaRocks version, for installation issues (run luarocks --version):

[kingluo]

This bug is due to the iteration of boolean items.

To avoid moving table (self.values) items frequently, we append the new items only to the table.

apisix/apisix/core/config_etcd.lua

Line 478 in 69df734

self.values[pre_index] = false

When the size of the table exceeds 100, the obsoleted items will be cleared. Before the clear, if the compacted event happens, it will iterate the self.values to invoke all clean handlers, which involves the false item and break the type assertion.

apisix/apisix/core/config_etcd.lua

Line 376 in 69df734

config_util.fire_all_clean_handlers(val)

So the bugfix is to skip the false items.

BTW, in the new watch mechanism, no compacted event happens anymore, because we watch the whole apisix keyspace, not individual watching for each admin resource. So this bug will be not triggered anymore.

[lingsamuel]

should be fixed by #8493 (already in 3.2.1) and #9811