Logbook July 2026

Logbook — July 2026

Week of Jun 29 — A single Saturday entry: opening the migration of the docs and auto-approve workflows onto the nix-enabled ARC runners, continuing the CI fleet move begun in late June.

Saturday 4th

CI runner migration (PV)

• Opened ci: migrate docs and auto-approve workflows to nix-enabled-runners #5308 — move the docs and auto-approve workflows onto the nix-enabled ARC runner fleet

Week of Jul 13 — Unit-test health week. cardano-wallet-unit was running the CI shards out of memory and, once that was bounded, the shards themselves turned out to be matching zero examples. Wednesday bounded the unit suite's memory — bounded applyNM list properties, deduplicated assertions, and restored sane RTS defaults. Thursday landed the memory fix, migrated the docs and auto-approve workflows onto nix-enabled runners, and repaired the unit-test fixtures behind a prose metadata error message (balance-tx golden regenerated); a throwaway PR confirmed the shard match patterns had been running no examples. Friday revived the wallet-unit CI shards properly — run from lib/unit, single-capability, with cardano-node provided — hardened the privileged-ports test, stopped echoing raw user JSON in metadata errors, and normalized the pool DB property harness to own its DBLayer per evaluation.

Wednesday 15th

Bound unit-test memory (PV)

• Opened fix: bound cardano-wallet-unit memory use #5310 — bound applyNM list properties, deduplicate property assertions, and restore bounded RTS defaults so the unit suite stops exhausting CI memory

Thursday 16th

Memory fix + fixture repair + CI runners (PV)

• Merged fix: bound cardano-wallet-unit memory use #5310 — bounded generation for the unit suite
• Merged ci: migrate docs and auto-approve workflows to nix-enabled-runners #5308 — docs and auto-approve workflows now run on the nix-enabled ARC fleet
• Closed fix(ci): correct unit-test shard match patterns (were running 0 examples) #5311 — throwaway that surfaced the shards were matching zero examples, superseded by the shard revival
• Repaired the unit-test fixtures behind a prose metadata error message and regenerated the ApiBalanceTransactionPostData testnet golden

Friday 17th

Revive wallet-unit CI shards (PV)

• Merged fix: repair unit-test fixtures and revive the wallet-unit CI shards #5312 — run the wallet-unit shards from lib/unit single-capability with cardano-node provided, key the privileged-ports test off the bind result, and stop echoing raw user JSON/keys in metadata error messages
• Merged Normalize pool DB property test harness (own the DBLayer per evaluation) #5315 — own the DBLayer per evaluation to fix a flaky pool DB SQLITE_MISUSE and margin round-trip

Week of Jul 20 — CI-fleet and release-unblocking week. The nightly Release had been dead for days, stuck queued forever against a stale self-hosted runner label no online runner matched; the fix repoints every remaining linux job — release, verify, publish, windows, mithril-sync, local-cluster-stress — at the nix-enabled-runners fleet CI already uses. In parallel the Windows E2E gate, which timed out after ~2h on a Mithril + full-sync preprod run and blocked releases, is replaced by a fast local-cluster integration smoke against the cross-compiled Windows binaries (with named-pipe node-socket support). A zero-margin stake-pool retirement assertion that flaked the SHELLEY_STAKE_POOLS suite is wrapped in eventually, and migration-test coverage lands alongside the root-key storage v2 (Argon2id+PBKDF2) upgrade.

Tuesday 21st

Stake-pool test flakiness (PV)

Issue: #5317

• Merged test: retry pool-list check after epoch-3 retirement in STAKE_POOLS_JOIN_03 #5318 — wrap the post-waitForEpoch 3 pool-list assertion in eventually (matching its sibling tests) to fix the intermittent SHELLEY_STAKE_POOLS failures rooted in the one test-cluster pool that retires at epoch 3

Wednesday 22nd

CI runner migration to nix-enabled-runners (PV)

• Merged ci(release): run linux jobs on nix-enabled-runners (nightly release is stuck) #5316 — repoint the release.yml and linux-e2e.yml linux jobs at nix-enabled-runners so the nightly Release stops queueing indefinitely against the stale cardano-wallet runner label
• Merged ci: update all remaining linux workflow jobs to nix-enabled-runners #5320 — migrate the last stragglers (verify-release, publish-release, windows, windows-e2e, linux-mithril-sync, local-cluster-stress) off the dead runner label

Root-key storage v2 (PV)

• Contributed root-key migration test coverage and an integration httpManager export to feat: upgrade root-key storage to v2 (Argon2id+PBKDF2) #5294 — cover V1→V2 root-key migration over CLI and HTTP as the Argon2id+PBKDF2 storage upgrade lands

Thursday 23rd

Windows E2E: local-cluster smoke (PV)

• Merged ci(windows): local-cluster smoke instead of preprod e2e #5321 — replace the ~2h Windows preprod Mithril + sync E2E (which timed out at ~57% and blocked releases) with a fast local-cluster integration smoke — a single TRANS_CREATE_01x transaction against the cross-compiled Windows wallet/node/cli, including Windows named-pipe node-socket support