tag: Compute-Engine · 6 items
- Platform/SRE — Learn: New console visibility into MIG instance distribution across zones and machine types — useful context for capacity planning and troubleshooting, but no operational change required.
- CI/CD — Skip
- Leader — Skip
- Platform/SRE — Plan: Two items worth acting on this quarter: enable graceful shutdown on production Compute Engine instances where long-running tasks risk data loss on stop/delete, and review the Cloud Armor WAF v33-stable and v3.0-stable ruleset changes to assess whether updated detection rules alter behavior for existing tuned deployments.
- CI/CD — Skip
- Leader — Skip
- Signals: deprecation/EOL deadline mentioned: October 19, 2026 · GA announcement
- Platform/SRE — Act: P100 GPU instances are already shut down as of September 15, 2026 — audit immediately for any affected workloads. T4 and P4 GPUs reach EOS on August 1, 2027; begin migration planning to G2 (L4) or G4 (RTX PRO 6000) machine series before that date.
- CI/CD — Skip
- Leader — Plan: T4 and P4 GPU EOL on August 1, 2027 blocks new 3-year CUD purchases and renewals for those SKUs; review any committed GPU spend and revise infrastructure roadmap to target supported machine series (G2/G4) within this planning cycle.
- Signals: deprecation/EOL deadline mentioned: September 22, 2026; September 15, 2026; August 1, 2027
- Platform/SRE — Plan: Cloud SQL (MySQL and PostgreSQL) now auto-enables PITR after DR switchover/failover asynchronously, reducing RTO — review your DR runbooks to reflect the faster recovery path. The GA Z4D machine series (AMD EPYC Turin, up to 3 TB local storage) is worth evaluating this quarter for storage-heavy workloads like databases, search, or vector stores.
- CI/CD — Skip
- Leader — Learn: The Z4D GA instances and Cloud SQL PITR improvements are worth noting for teams standardizing on GCP for storage-heavy or analytics workloads, but neither forces a strategic decision or pricing change now.
- Signals: GA announcement
- Platform/SRE — Plan: The M4N machine series is now GA on GCP, offering up to 400 Gbps network and 1M IOPS for memory/network-intensive workloads like vector databases and RAG layers — evaluate whether it fits high-memory workload placements this quarter. CVE-2026-12710 in Application Integration was already patched server-side on April 4, 2026; no customer action required.
- CI/CD — Skip
- Leader — Learn: GCP’s M4N instance family (GA) targets high-memory AI infrastructure workloads such as vector databases and in-memory RAG layers — relevant context for future GCP AI/ML platform architecture discussions, but no immediate strategic or budget decision is forced.
- Signals: GA announcement · CVE-2026-12710 — CISA KEV: not listed, EPSS n/a
- Platform/SRE — Learn: Advanced Compute Images (Preview) offer pre-tuned AI/ML/HPC OS images with drivers and Slurm pre-installed, potentially simplifying GPU node provisioning; separately, BigQuery hybrid search has been temporarily disabled — check if any workloads depend on the VECTOR_SEARCH hybrid mode.
- CI/CD — Skip
- Leader — Skip