It is a problem with SQL, after so long use the share table reached its all time limit. This is not related to 5 hour round, but to total pool uptime (starting from december). Those shares were not counted in database in fact. Fortunately all shares are also counted separately on each backend, so I only need to merge this data from backends now.