Docs, ticks, working app

2026-09-05 15:05:39 +01:00
parent 41362aa3d8
commit 51e577f01e
41 changed files with 1039 additions and 683 deletions
@@ -0,0 +1,39 @@
+# 020 — File input for ingestion, with restart-safe tailing
+
+## Summary
+
+`ingest.rs` only reads stdin today. The wishlist asks for an optional file path as
+an alternative input source, plus resilience to the log file being recreated (e.g.
+the host process being tailed restarts and truncates/recreates its log file) —
+ottyl should keep tailing rather than treating that as EOF-and-done.
+
+## Scope
+
+- New `clap` arg (e.g. `--file <path>`), mutually exclusive-in-spirit with stdin:
+  when given, ingest reads from the file instead of stdin; when absent, behavior is
+  unchanged (stdin, as today).
+- File-mode ingestion must behave like `tail -f`: keep reading as new lines are
+  appended, don't exit on reaching current EOF.
+- Detect file recreation (the common "log rotated / process restarted and reopened
+  the path" case) — e.g. by tracking the inode/file id and re-opening when it
+  changes, or noticing truncation (current read offset > file size) — and resume
+  tailing the new file from its start rather than exiting or getting stuck.
+- Sequence numbers continue incrementing across a recreation event (per
+  `IndexKey::IngestedLog`, sequence is ottyl's own line counter, not tied to the
+  source file).
+- Reuse the existing `IngestedLine` / `EngineMessage::Ingest` plumbing — this only
+  changes where lines come from, not how they're indexed.
+
+## Acceptance criteria
+
+- `ottyl --file some.log` ingests existing content and then picks up appended
+  lines live.
+- Deleting and recreating the file at the same path (simulating a host restart)
+  does not kill ottyl or stop ingestion — new content from the recreated file
+  continues to be ingested.
+- Running with no `--file` is unchanged: stdin ingestion as today.
+
+## Depends on
+
+None — parallels the existing `ingest::run`, doesn't require any of the metrics/
+filter machinery.