bryancall opened a new issue, #12944:
URL: https://github.com/apache/trafficserver/issues/12944
## Problem
ATS currently has **four separate log files** with inconsistent timestamp
formats and overlapping purposes, while comparable proxy servers (Nginx, Envoy,
HAProxy) use only one or two.
### Current log files
| Log File | Written By | Timestamp Format | Purpose |
|----------|-----------|-----------------|---------|
| `squid.log` | Logging subsystem (`LogObject`) | epoch.ms (e.g.
`1741210395.123`) | Per-transaction access log |
| `error.log` | `Log::error()` via `LogObject::va_log()` |
`%Y%m%d.%Hh%Mm%Ss` (e.g. `20260305.21h33m15s`) | Connection failures, DNS
errors |
| `diags.log` | `Diags::print_va()` | `[DiagTimestamp]` with thread name +
level | Debug messages, warnings, notes, errors |
| `traffic.out` | Same `Diags` system (stdout/stderr) | Same as `diags.log`
| Duplicate of diags.log to stdout |
### Comparison with other proxies
| Proxy | Log Files |
|-------|-----------|
| Nginx | 2 — `access.log` + `error.log` |
| Envoy | 1 — access log (errors to stderr) |
| HAProxy | 2 — access log + admin log |
| **ATS** | **4** — `squid.log` + `error.log` + `diags.log` + `traffic.out` |
## Issues
1. **Three different timestamp formats** across log files make cross-log
correlation difficult
2. **`diags.log` and `traffic.out`** are the same Diags system writing to
two destinations — redundant
3. **`error.log` and `diags.log`** both contain error-level messages but use
completely different code paths (`Log::error()` vs `Diags` Error level) with
different formats
4. **No SM-aware error macros** — `SMDbg`/`TxnDbg` prefix debug messages
with `[SM_ID]` but there's no `SMError`/`TxnError` equivalent for
`Log::error()`, `Log::warning()`, or `Log::note()`. This makes it impossible to
correlate error.log entries back to a specific transaction without manually
adding the SM ID (as seen in PR #12932)
5. **"Squid log" is a legacy name** — the default format hasn't been actual
Squid format in a long time, but field names (`client_req_timestamp_squid`,
`cqtq`) and config still reference "squid"
## Proposed Changes
### 1. Add SM-aware error/warning/note macros
```cpp
#define SMError(fmt, ...) Log::error("[%" PRId64 "] " fmt, sm_id,
##__VA_ARGS__)
#define SMWarning(fmt, ...) Log::warning("[%" PRId64 "] " fmt, sm_id,
##__VA_ARGS__)
#define SMNote(fmt, ...) Log::note("[%" PRId64 "] " fmt, sm_id,
##__VA_ARGS__)
// And TxnError/TxnWarning/TxnNote for HttpTransact context
```
### 2. Standardize timestamp format
Pick one format and use it everywhere, or at minimum make them correlatable
(e.g. all ISO 8601).
### 3. Rename "squid log" to "access log"
Update config names, field names, documentation, and code references.
### 4. Consolidate log files
Long-term goal: merge `error.log`, `diags.log`, and `traffic.out` into a
single diagnostic/error log, similar to how Nginx has one `error.log` for
everything that isn't per-transaction access logging.
## Notes
- Items 1-3 can be done incrementally without breaking changes
- Item 4 (log consolidation) is a larger architectural change that should be
designed carefully
- The SM-aware macros (item 1) are the most immediately useful and could
unblock PR #12932's approach of adding `sm_id` to error log messages in a
standardized way
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
