d3v-null

facetselfcal on MWA data with the sp5505 container

This guide runs facetselfcal (lofar_facet_selfcal, R. van Weeren) on the same MWA observation as the rapthor guide: obs 1224424640, the EoR0 field converted from a hyperdrive uvfits. It uses the same sky model and solve settings as the rapthor check there, so the two can be compared directly.

The image

d3v-null

Whole-night MWA EoR1 power spectra: which RFI flagging helps

How to reproduce the flagging comparison and the full-night measurement set for one MWA EoR1 high-band night (2026-09-05 UTC, 71 × 2-min obs 1472670968–1472679368, MWAX correlator, 128 tiles, coarse channels 131–154 = 167–198 MHz).

Result. With DI-calibrated data and CHIPS half-band power spectra (lower 167.0–182.4, middle 174.7–190.1, upper 182.4–197.8 MHz):

• No flags (input flags only) and a gentle whole-band bmetrics mask give the cleanest spectra.
• AOFlagger, even with a relaxed full-band strategy that adds only 0.17% of cells, raises Δ² at k ≈ 0.6–0.8 h/Mpc by ~8–10× (lower), ~3–5× (middle) and ~10–15× (upper half-band), by ~8–10× at k ≈ 0.28 in the upper band, and at k > 2 by 3–4× / ~2.5× / ~1.6–2× (lower / middle / upper). Its scattered, baseline-dependent flags imprint spectral structure.

d3v-null

How the Birli UVW-change plot was made

This plot is from MWATelescope/Birli#203. It shows how much the UVWs Birli writes change between Marlu 0.19.0, which agrees with cotter, and the unreleased precession on Marlu main (e739762: rotate by the apparent sidereal time; 150623a: take the phase centre's aberration out with a rigid rotation). The same changes are on feat/conventions, via Marlu#43.

Steps

The data are Birli's test files for 1254670392 (tests/data/1254670392_avg), using 2 coarse channels (gpubox 23 and 24) and 2 timesteps. Both Birli builds need aoflagger (apt-get install aoflagger-dev) for --no-rfi.

d3v-null

CephFS returns ENOSPC while df reports 150 GB free

Rook-Ceph on Oracle OKE, size=1 CephFS pool, mixed 100/300 GiB OSDs across three unequal hosts. A rapthor/Toil workload died mid-run with OSError: [Errno 28] No space left on device while every filesystem-level check said there was plenty of room. Writing this up because the symptom is actively misleading and the usual first check (df) is the wrong tool.

Symptom

d3v-null

IDG + WSClean on AMD GPUs (ROCm) on Pawsey Setonix

Findings from building and testing ASTRON's experimental HIP/ROCm port of the Image Domain Gridder (IDG) on Setonix, 14 Oct 2026.

TL;DR

• IDG has no official ROCm support. The only HIP port is ASTRON's llu/hip-rocm branch (Dec 2021, forked from a pre-1.0 snapshot, ~460 commits behind main). It builds against ROCm 6.3.0 with a few CMake shims and runs on the MI250X.
• A spack recipe for it (idg@hip-rocm +rocm) and a wsclean +rocm variant live on the branch idg-rocm of d3v-null/pawsey-spack-config (PR to PawseySC pending).
• Degridding (predict) on the GPU is correct. Gridding on the GPU is not: Stokes-I dirty and model images are a constant 1.387× brighter than the CPU result. Do not use -idg-mode hybrid for imaging until that is fixed; it is fine for -predict.
• The GPU path is not faster for small problems: each wsclean run compiles 36 HIP kernels at star

d3v-null

d3v-null

Execution Framework components: requirements gap fill

Generated 2026-10-09 (read-only; nothing was applied, synced or changed in any cluster or repo). Secret values are never printed; only Secret names, ExternalSecret Vault paths and key names.

Sources actually used

Ref	Source	Notes
dump	Artifacts of GitLab job 17034588345 (project ska-telescope/src/src-api/ska-src-api-deployment-stack), out/deploy-ci-2927014939/cluster-dump//*.json, out/deploy-ci-2927014939/log, report.html	Job 17048104007 had no artifacts at 2026-10-09 03:23 UTC: the API returned 404 Not Found for /jobs/17048104007/artifacts and /jobs/17048104007 reported "status":"running" (started 2026-10-09T02:36:00Z). The fallback job was used.

d3v-null

BEIGE-49: Testing pathfinder against the SRCNet deployment stack

Short answers to the open questions on pathfinder API and integration tests. Links are pinned to commits so the line numbers stay valid.

Repos referenced: pathfinder · deployment stack · services-cd · auth-api · rucio-deployment

d3v-null

Fixing the Fornax A sky model for MWA EoR1 calibration

The Fornax A entry in srclist_pumav3_EoR0LoBES_EoR1pietro_CenA-GP_2023-11-07 over-predicts visibility amplitude on the baselines MWA calibration uses. Fornax A carries about 46% of the di-calibrate weight at --uvw-min 30l, so that error sets the flux scale for the whole field.

Multiplying the seven Fornax A components by about 0.8 and changing nothing else:

	original model	rescaled

d3v-null

Rapthor on MWA data at Pawsey Setonix with the rapthor-mwa container

ghcr.io/d3v-null/rapthor-mwa is a lean (no Jupyter/conda) Spack build of the LOFAR Rapthor direction-dependent calibration and imaging pipeline with MWA support compiled in:

component	version	MWA notes
rapthor	2.1.20260630 (py-rapthor +mwa)	patched: DP3/WSClean pointed at the MWA beam, facet-beam options dropped, flagged-tile and solar-model robustness fixes (list below)
DP3	6.6.20260819	applybeam.coefficients_path / solve.coefficients_path = /opt/mwa_full_embedded_element_pattern.h5