Fix temp-file race between concurrent clickhouse_import_support scripts

[jamesqo]

Problem

write_clickhouse_config_file names its temp config file with a second-resolution timestamp in $(pwd):

clickhouse_client_config_$(date "+%Y-%m-%d-%H-%M-%S").yaml

Every script that sources clickhouse_client_command_line_functions.sh from the same working directory (e.g. verify-management-state.sh and the get_database_currently_in_production.sh child it spawns, both running with cwd = the SSH user's home dir) generates the same filename when they land in the same second. One process's creation (rm -f + touch) or shutdown cleanup then deletes the file out from under the other, and clickhouse-client fails with:

Code: 107. DB::Exception: Configuration file /home/cbioportal_importer/clickhouse_client_config_2026-08-26-09-15-02.yaml doesn't exist and there is no embedded config. (FILE_DOESNT_EXIST)

The caller then misreads this as "database does not exist" and aborts. Observed failing import_triage_dag / verify_management_state on 2026-08-26 (the error timestamp matches the colliding filename's second exactly). A stray zero-byte clickhouse_client_config_2026-07-22-*.yaml in the same home dir is debris from an earlier instance of the same collision.

ccclf_database_exists.txt / ccclf_database_table_list.txt are fixed filenames in $(pwd) with the same cross-process collision, without even needing same-second timing.

Fix

Name all three temp files by process id instead — concurrent processes cannot share a PID, so no process can collide with (or clean up) another's files. Creation, permissions (600), content checks, and shutdown cleanup are unchanged.

🤖 Generated with Claude Code