Bulk deck export: manifest reconciliation, replayable failures, and collision-proof naming are the SLA backbone
`batch-vsdx-pptx` converts libraries of .vsdx into standardized slide assets. Silent partial success, clobbered filenames, and non-rerunnable rows are what blow SLAs. Freeze manifest columns—source path, repo version, output name, status code, duration, hash—and bucket files with data links or huge embeds into their own queues. Reconcile successful row counts against object-store counts daily; every delta needs a ticket, not a “looks fine” folder glance.
Bulk export workflow: golden + corrupt pilots define concurrency, then checkpoint manifests for resume
- Sample the largest, most complex, and a few known-bad files per directory; have data owners sign off manifest columns and naming—avoid ad-hoc CSV schemas per team that break automation joins.
- Flush manifest and failure lists every N rows for resume; outputs must include repo id and path hash; spot-check several PPTX files against the standard master to catch systematic font or safe-area drift before widening load.
- Archive manifest hashes beside read-only bucket snapshots; reruns skip rows flagged `manual_override` unless a forced change order says otherwise, protecting legally approved outbound decks.
VSDX to PPTX (batch) FAQ: manifests, resume, overwrites, bad buckets, large-file queues
Morning counts show fewer PPTX than source .vsdx—which manifest fields do we read before blaming random converter bugs?
Inspect per-row status, skip reasons, and whether an output URI exists—often case sensitivity, locks, or manual deletes, not flaky engines.
Multiple CI branches write one `out/` bucket—how do auditors tie artifacts to commits?
Manifests must log git SHA, source checksum, and artifact hash; never rely on unlabeled `latest` folders as the sole source of truth.
A few OLE-heavy diagrams stall the whole batch—isolate or globally throttle?
Spin up a heavy-embed queue with longer timeouts and lower parallelism; global throttling only stretches wait times for healthy files.
Reruns risk overwriting sales-beautified decks—how do we prevent a second fight?
Mark `manual_override` rows and change suffixes or subfolders; default reruns skip unless a ticket forces overwrite.
Compliance wants a point-in-time snapshot—is CSV enough?
No—also freeze bucket versions or read-only snapshots and hash the manifest file; spreadsheets alone can be edited after the fact.