llama 3.3 70b q4 generates valid terraform but wrong resource dependencies past 26k
tested at 26.4k context yesterday generating terraform for aws infrastructure. syntax is perfect but dependency order is completely backwards - like it creates route53 records before the load balancer exists, or references security group ids before they're defined. terraform validate passes but apply just fails immediately. anyone else seeing this?
hit this at 27.3k yesterday generating terraform for aws ecs service. generated valid hcl but had aws_ecs_service depending on aws_lb_target_group before the target group resource even existed in the config. terraform plan failed immediately bc dependency graph was completely backwards. syntax was perfect, resource blocks looked totally plausible, but execution order would have been a disaster if terraform hadn't caught it at plan time. makes me wonder how many other iac tools would just try to execute it anyway. what context size did you hit this at and was it also aws resources or other providers?
does it break on all resource types or just specific ones like target groups
imo terraform dependency breakage past 26k is worse than yaml or json failures bc terraform will actually execute the broken config and provision resources in the wrong order, potentially leaving your infra in an inconsistent state. could be wrong but wrong flag order in docker is annoying, wrong resource dependencies in tf can cost real money