-- Test gprecoverseg from config file uses the correct dbid.
--
-- In github issue 9837 dbid in gp_segment_configuration is not
-- consistent with dbid in file internal.auto.conf.
-- This is caused by gprecoverseg fetch the smallest dbid in
-- gp_segment_configuration which is not occupied by others when
-- adding a new mirror. When dbid in gp_segment_configuration is not
-- continous, the inconsistent issue will happen

include: helpers/server_helpers.sql;
CREATE

--
-- generate_recover_config_file:
--   generate config file used by recoverseg -i
--
create or replace function generate_recover_config_file(datadir text, port text) returns void as $$ import io import os myhost = os.uname()[1] inplaceConfig = myhost + '|' + port + '|' + datadir configStr = inplaceConfig + ' ' + inplaceConfig  f = open("/tmp/recover_config_file", "w") f.write(configStr) f.close() $$ language plpythonu;
CREATE

SELECT dbid, role, preferred_role, content, mode, status FROM gp_segment_configuration order by dbid;
 dbid | role | preferred_role | content | mode | status 
------+------+----------------+---------+------+--------
 1    | p    | p              | -1      | n    | u      
 2    | p    | p              | 0       | s    | u      
 3    | p    | p              | 1       | s    | u      
 4    | p    | p              | 2       | s    | u      
 5    | m    | m              | 0       | s    | u      
 6    | m    | m              | 1       | s    | u      
 7    | m    | m              | 2       | s    | u      
 8    | m    | m              | -1      | s    | u      
(8 rows)
-- stop a primary in order to trigger a mirror promotion
select pg_ctl((select datadir from gp_segment_configuration c where c.role='p' and c.content=1), 'stop');
 pg_ctl 
--------
 OK     
(1 row)

-- trigger failover
select gp_request_fts_probe_scan();
 gp_request_fts_probe_scan 
---------------------------
 t                         
(1 row)

-- wait for content 1 (earlier mirror, now primary) to finish the promotion
1U: select 1;
 ?column? 
----------
 1        
(1 row)
-- Quit this utility mode session, as need to start fresh one below
1Uq: ... <quitting>

-- make the dbid in gp_segment_configuration not continuous
-- dbid=2 corresponds to content id =0
set allow_system_table_mods to true;
SET
update gp_segment_configuration set dbid=9 where dbid=2;
UPDATE 1

-- trigger failover
select gp_request_fts_probe_scan();
 gp_request_fts_probe_scan 
---------------------------
 t                         
(1 row)

-- wait for content 0 (earlier mirror, now primary) to finish the promotion
0U: select 1;
 ?column? 
----------
 1        
(1 row)
-- Quit this utility mode session, as need to start fresh one below
0Uq: ... <quitting>

-- generate recover config file
select generate_recover_config_file( (select datadir from gp_segment_configuration c where c.role='m' and c.content=1), (select port from gp_segment_configuration c where c.role='m' and c.content=1)::text);
 generate_recover_config_file 
------------------------------
                              
(1 row)

-- recover from config file, only seg with content=1 will be recovered
!\retcode gprecoverseg -a -i /tmp/recover_config_file;
-- start_ignore
-- end_ignore
(exited with code 0)

-- after gprecoverseg -i, the down segemnt should be up
-- in mirror mode
select status from gp_segment_configuration where role='m' and content=1;
 status 
--------
 u      
(1 row)

-- recover should reuse the old dbid and not occupy dbid=2
select dbid from gp_segment_configuration where dbid=2;
 dbid 
------
(0 rows)

update gp_segment_configuration set dbid=2 where dbid=9;
UPDATE 1
set allow_system_table_mods to false;
SET

-- we manually change dbid from 2 to 9, which casue the
-- corresponding segment down as well, so recovery full
-- at here
!\retcode gprecoverseg -aF;
-- start_ignore
-- end_ignore
(exited with code 0)

-- rebalance the cluster
!\retcode gprecoverseg -ar;
-- start_ignore
-- end_ignore
(exited with code 0)

-- remove the config file
!\retcode rm /tmp/recover_config_file