Building Durable Workflows on `Postgres`

Checkpointing state in the database lets jobs resume after crashes. It cuts orchestration overhead when a small product already runs on Postgres, but complex branching still needs care.

[ KEY POINTS ]

Durable workflows persist completed steps as checkpoints, so a crash restarts from the last finished step instead of rerunning everything.
External orchestrators like Temporal, Airflow, and AWS Step Functions add a central coordinator plus storage; that raises operational weight.
A Postgres-based design fits small SaaS backends already using the DB, especially for billing, emails, imports, and LLM batch jobs.

Originalnews.hada.io/topic?id=29999Read original →

// related

#0001
#0001Infra & SaaS GeekNews13 hours ago
Durable Workflows Can Be Built on `SQLite` Alone
50radar
ObeliskWorkflow runtime — replays state from execution logs
Durability comes from saved workflow state, not heavy infrastructure. If progress is replayable, SQLite is enough for small reliable job orchestration.
- The core requirement is persisted progress: reruns and recovery work once each workflow step leaves state behind.
- Obelisk stores workflow progress in an execution log and replays from durable history, matching activity retry patterns.
- This fits background jobs, billing retries, and AI task pipelines before adopting heavier queues or workflow platforms.
Source: news.hada.io/topic?id=30014Read original →
FIG-0011:1
50radar
FIG-0011:1
#0002
#0002Infra & SaaS GeekNews15 hours ago
`Shopify` Replaces Redis Inventory Reservations With MySQL
50radar
ShopifyE-commerce SaaS — unified checkout and store ops
Oversell prevention moved from cache-style reservations to transactional rows using SKIP LOCKED. Worth copying only when checkout correctness beats simple counter performance.
- MySQL 8 SKIP LOCKED lets workers claim available reservation rows without blocking the whole queue; useful for high-contention checkout paths.
- The redesign uses one row per sellable unit, not a quantity column. It trades table size for clearer locking and recovery semantics.
- The pattern fits limited-stock sales, tickets, and booking flows. For tiny catalogs, a simpler counter plus idempotent order checks is enough.
Source: news.hada.io/topic?id=30006Read original →
FIG-0021:1
50radar
FIG-0021:1
#0003
#0003Infra & SaaS GeekNews15 hours ago
How Container Registries Work: Push/Pull Images Directly
40radar
Registry bugs often hide below the docker push surface. Understanding blobs, tags, manifests, and platform metadata helps debug broken deploys faster.
- A registry is a content-addressable blob store; layers are referenced by digest, so missing blobs break pulls even when tags look valid.
- Tags are mutable pointers, not versions. Pinning image digests is safer for Vercel-style build pipelines and reproducible deploys.
- Platform mismatches come from manifest metadata. Multi-arch images need explicit checks before shipping ARM/local builds to x86 servers.
Source: news.hada.io/topic?id=30004Read original →
FIG-0031:1
40radar
FIG-0031:1

Building Durable Workflows on `Postgres`

// related

Durable Workflows Can Be Built on `SQLite` Alone

`Shopify` Replaces Redis Inventory Reservations With MySQL

How Container Registries Work: Push/Pull Images Directly