F2: Execute had no status guard, so a redelivered or stale message reran a
finished run. Runs that are neither queued nor running are logged and acked
without work; "running" still executes so a worker that crashed mid-run
recovers.
F4: failRun, finishNodeFailed and MarkRunFailed wrote the terminal status on
the caller's context, which is exactly the context that is already cancelled
when the run failed because of a shutdown or a sibling node — the write was
refused and the run stayed stuck in "running". They now write on a
context.WithoutCancel copy with a 10s timeout; the consumer's best-effort
MarkRunFailed does the same.
F5: peak memory scaled with the host's core count. LP_MAX_WORKERS (default 8)
is now the total window-worker budget of a run and LP_MAX_NODES (default 2) the
node fan-out; each node gets max(1, workers/nodes) through the new
landscape.ComputeWithWorkers. Compute and service.ComputeFunc keep their
signatures: Compute delegates with MaxWorkersPerNode, and a legacy ComputeFunc
handed to NewLandscapeService is wrapped.
F6: the stage loop used a plain errgroup, so a failing node let its siblings
run to completion. It now uses errgroup.WithContext and passes the group
context into runNode, cancelling siblings mid-compute.
F7: wrap every float32 product in an explicit conversion so the compiler
cannot fuse it into an FMA, keeping the kernels bit-identical to numpy's
per-op rounding.
F8: README's builder package list now matches the Dockerfile (adds g++), and
the Landscape section documents the memory model and the two new knobs.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
F1: ValidateFileContent rejected the pipeline's own output. GDAL writes a
BigTIFF header ("II+\0" / "MM\0+") whenever BIGTIFF=IF_SAFER kicks in for a
large raster, so accept both alongside the classic TIFF magic and reword the
error to "file is not a valid TIFF/BigTIFF GeoTIFF". GodalIO.Create now
delegates to an unexported createGTiff so a test can force BIGTIFF=YES and
assert the produced header passes the check.
F3: a user-supplied filename went straight into the S3 object key for dataset
uploads and for both dataset and dataset-group SLD uploads, so "../../x.sld"
escaped the dataset's own prefix. Reduce every such name to its base with a
shared safeObjectName helper, rejecting names with no usable base.
F9: rename the shadowing inner "raw" in the PATCH dg_id case to "s".
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
POST /landscape/runs documents an optional body, but DecodeJSON returned
io.EOF on an empty one and the handler turned that into 400 — breaking the
most common call, an all-defaults run. decodeRunRequest now treats io.EOF as
the zero request; a malformed body is still 400.
The Dockerfile's Alpine ABI coupling was enforced only by a comment: a
floating golang:1.26-alpine builder against a pinned alpine:3.24 runtime
would build green and die at startup on a libgdal soname mismatch as soon as
the Go image rolls forward. Both stages are now pinned to 3.24.
Also: the route-mount test walks the router so all six routes are asserted
(GET /slots and GET /results included), and the spec documents the 404 those
two return when migration 00011's seeded groups are absent.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Adds the /landscape routes (slots, slot upload, runs, run detail, results),
mounts them in the router, wires the landscape repository, job publisher,
godal raster IO and Compute into the composition root, and registers the
gis-landscape-runner consumer in the worker.
Also fixes the Dockerfile so the image actually runs: godal's cgo needs g++,
and the runtime base must match the builder's Alpine release or libgdal's
soname does not resolve.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Review round 1: size per-worker buffers off the largest actual window
instead of WindowSize^2 (avoids allocating ~200+ MiB per worker for
small/partial rasters), cap workers at 8 regardless of GOMAXPROCS, remove
partially-written outputs when a run fails, wrap input-open errors with
node/group context, guard checkGrids against a node with no inputs, and
clamp a non-positive WindowSize so raster.Windows can't loop forever.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
