Release gstop v1.5.0 - #1

What changed

• bump gstop to v1.5.0
• add the health dashboard with SQL/plan details, plan-change history, statistics/index checks, wait events, and DB CPU
• make collection cancellable and enforce per-module timeouts so refresh and exit are not blocked by slow queries
• render emergency scenarios as a dynamic bottom overlay on short terminals
• add gsbench with TP/AP/mixed CPU, connection pool, thread pool, dynamic memory, plan regression, lock, and vacuum scenarios
• fix openGauss thread-pool session counting by joining pg_stat_activity and pg_thread_wait_status on the unique (pid/tid, sessionid) identity

Root cause

The instance summary joined the two session views only by sessionid. With thread pooling enabled, both views contained 120 rows where sessionid=0, producing a 120 ×ばつ 120 many-to-many expansion. This inflated a real session count near 140 to more than 14,000 and falsely triggered the ConnectionsFull emergency scenario.

Impact

The CONNECTION value and ConnectionsFull detection now use the real session cardinality. On the reproduced og5 instance, the corrected join returned 141 rows instead of 14,421.

Validation

• go test ./...
• go test -race ./internal/...
• go vet ./...
• regression test for the composite session/thread join
• live og5 query comparison of activity rows, the old join, and the corrected join
• Linux ARM64/x86_64 release binaries built and SHA-256 verified locally