Process-level CPU limit
pg_pcpu_limit constrains selected database-process CPU use by suspending and resuming processes. It is not a workload resource-group or queueing framework.
OPERATIONS & OBSERVABILITY
Operate Halo with familiar controls and visible evidence.
Halo exposes settings, catalogs, statistics, reports, resource controls, and administration tools for deployment fit, workload control, historical diagnosis, and day-to-day inspection.
Fit the documented deployment environment
The manual lists a broad platform range. Treat it as documented coverage; confirm exact versions, packages, and current support status for the intended release.
Processor familiesx86 · ARM · MIPS · Alpha
Documented x86 families include Intel, AMD, Zhaoxin, and Hygon; ARM includes Kunpeng and Phytium; MIPS includes Loongson; Alpha includes Sunway.
Operating-system familiesLinux and FreeBSD
Kylin, NeoKylin, UnionTech, openEuler, Ubuntu, Red Hat Enterprise Linux, CentOS, and FreeBSD are listed in the installation material.
Installer workflowAutomated preparation
The installer detects the operating system and covers dependency handling, service-user setup, cluster initialization, core configuration, archive setup, licensing, service startup, compatibility-mode databases, and HugePages handling.
Character setsUTF-8 and GB18030
UTF-8 is the documented default. The manual states support for GB18030-2022 Level 3. GBK is not promoted because its status is not stated with the same clarity.
Tune through defined operating concerns
Connectivity and authentication
Network listeners, client access, authentication rules, connection protection, and session behavior define how workloads enter the database.
Memory, I/O, and processes
Memory allocation, background processes, parallel work, disk cost assumptions, and I/O behavior adapt the engine to the host and workload.
WAL, archive, and durability
WAL generation, checkpoints, commit behavior, archive handling, and pglz or lz4 compression choices balance throughput, recoverability, and storage.
Query optimization
Planner switches, cost assumptions, statistics, statement behavior, and parallel-query controls influence plan selection. They are tuning inputs, not universal presets.
Autovacuum and maintenance
Automatic cleanup and analysis protect table health, visibility information, and planner statistics over time.
Apply focused resource guardrails
Per-role connection limits
Connection ceilings can keep one database identity from consuming all session capacity.
Idle-session timeout
Inactive sessions can be closed after a defined interval, releasing abandoned connections and related session resources.
Catalogs and views expose current state
- 64base catalogs
- Database objects, security, optimizer statistics, text search, federation, and replication metadata.
- 11Halo catalogs
- Compatibility objects, query mapping, and model/runtime registration.
- 64system views
- Activity, I/O, WAL, replication, SSL, progress, and object statistics.
- 139documented system objects
- Presented by operational domain, not as a manual-sized list.
Operational visibility by domain
Sessions, activity, and locks
Current queries, backend activity, locks, cursors, prepared statements, two-phase work, role state, and row-lock inspection support live investigation.
I/O, memory, and storage
Shared buffers, backend memory, I/O, background writing, checkpoints, free-space maps, tuple and index bloat, relation sizes, and low-level pages support progressively deeper diagnosis.
WAL, archive, and replication
Archive, WAL, sender/receiver, slots, origins, subscriptions, errors, synchronization state, lag, and recovery prefetch expose continuity health.
Maintenance progress
Progress is available for analyze, base backup, cluster or VACUUM FULL, copy, index creation or reindexing, and vacuum operations.
Configuration and connection security
Settings, configuration-file parsing, HBA and ident checks, SSL state, certificate details, extensions, and runtime versions support configuration review.
HWR moves from a live symptom to a historical interval
HWR collects scheduled database snapshots and produces HTML reports across a selected interval. It can collect sequentially from multiple databases and servers.
COLLECT
Scheduled snapshots
Activity, workload, I/O, WAL, function, and object statistics.
RETAIN
Historical samples
Configurable retention and sample metadata.
SELECT
Interval or baseline
Sample IDs · time range · baseline · latest pair.
REPORT
HTML evidence
Top-N activity, resource behavior, and optional growth views.
A report needs at least two samples. HWR depends on PL/pgSQL, dblink, scheduled cron work, and pg_stat_statements; sampling resets pg_stat_statements, so timing should be coordinated with other monitoring consumers.
Diagnostics, execution reuse, and DBA tooling
Workload and plan diagnostics
pg_stat_statements aggregates planning, execution, row, block, I/O, and WAL metrics. auto_explain can record plans for slow statements. pgstattuple, amcheck, pg_visibility, pg_buffercache, pageinspect, and related modules cover integrity, bloat, cache, and storage evidence with documented cost and consistency limits.
Statement Sharing
none, normal, and exact modes control reuse of statement processing to reduce repeated parse and compile work. The manual’s single-environment benchmark is not republished as a general result.
DBeaver
The cross-platform DBeaver client can browse databases, schemas, tables, views, columns, constraints, and indexes; view data; use the SQL workspace; and manage transaction behavior. Bundled no-install packaging remains release-dependent and is not promised here.
Operational safeguards
Some inspection functions are privileged, can scan large relations, or do not provide a perfectly consistent snapshot. Diagnostic depth should be matched to impact and access policy.