AWS Solutions Architect Professional (SAP) study notes
Contents
Completed this one on a wintry Sydney morning. Despite a lack of sleep, I felt much better prepared vs. DevOps Pro, and it showed in the grade, too:
Overall Score: 75%
Topic Level Scoring
- High Availability and Business Continuity: 72%
- Costing: 75%
- Deployment Management: 57%
- Network Design: 42%
- Data Storage: 72%
- Security: 85%
- Scalability & Elasticity: 90%
- Cloud Migration & Hybrid Architecture: 85%
Scratchpad
- ASG;
- CloudFormation, using YAML/JSON for solution architecture:
- Which Services are supported?;
- Templates vs Stacks;
- “Fn::getAtt”: [ “WebServerHost”, “PublicIp”];
- Chef/Puppet integration OK, or use bootstrapping for application-level dependencies/config;
- By default, automatic rollback on error (resource provisioning charges still apply);
- WaitCondition;
- Resource deletion policy;
- Resource update policy;
- IAM Role definition and assignment;
- VPC creation and customisation (and pretty much everything within a VPC);
- EIPs and private IPs;
- Multiple VPCs (for peering) within the same Account only;
- Create/update Route 53 hosted zones;
- CloudFront, web/RTMP, geo black-/white-list, SSL (SNI/dedicated), PPPD HTTP methods never cached (proxied to origin), CNAME?, Invalidation API, naked domain alias, dynamic content support (cookie), Origin Access Identity (OAI);
- CloudHSM, key GSM within a VPC, non fault-tolerant (needs to be clustered), accessible via peering;
- CloudSearch?;
- CloudTrail, from multiple Accounts to an S3 bucket;
- CloudWatch, indefinite by default, 14 days for alarm history;
- Data Pipeline, use on AWS (EC2/EMR)or on-premise, pipeline container, data node (end dest), activity, precondition, schedule;
- Direct Connect, 802.1q, can address both public/private environments by relying on an underlying VIV, sub-1/1/10 Gbps dark fibre to AWS vs establishing a VPN, CGW vs VGW, also configure BGP failover, US-only: 1 Direct Connect will work for all Regions;
- Directory Services, AD Connector (existing), Simple AD (new);
- DR as a part of BC (eg hardware/software failure, network/power outage, physical damage). Spectrum of options:
- Backup & Restore: (i) backup (to AWS), (ii) retention policies; and (iii) security measures; eg access policies, encryption.
- Pilot Light: (i) Pre-configure functional (eg app/web, database) servers as AMIs for various functions; (ii) fire drill; (iii) consider automation (via CloudFormation).
- Warm Standby: (i) Run apps in an ASG (and/or other infrastructure); and (ii) keep ’em up-to-date (eg patches, config files).
- Multisite/active-active: (i) Duplicate non-AWS environment; and (ii) Configure weighted routing (Route 53) to route traffic on-premise/AWS environments;
- DynamoDB (cross-Region replication);
- EBS point-in-time (eg using the CLI), attached to a single EC2 only;
- EC2;
- D (dense storage);
- I (IOPS);
- R (RAM);
- T (t2.micro);
- M (main choice);
- C (compute);
- G (graphics);
- F (FPGA);
- P (mining);
- X (HANA/Spark);
- ECS?;
- EFS?;
- ElasticCache, Memcached (scale out, multi-threaded) vs Redis (scale up, persistency, multi-AZ);
- Elastic Beanstalk (EB), relatively simple (vs CloudFormation);
- Applications vs environments;
- Supported languages:
- Docker (single-/multi-container);
- Go 1.6;
- Java w/ Tomcat;
- Java SE (7, 8);
- .NET (IIS 7.5+);
- Node.js;
- PHP;
- Python 2.6+;
- Ruby 1.9+;
- Supported AWS services include:
- CloudWatch;
- IAM;
- RDS;
- S3;
- VPC (within a Region only);
- Elastic Transcoder?;
- ELB, CLB (single/multiple AZs, health checks, associate SGs, SSL offload, sticky sessions, IPv{4,6}, CloudWatch metrics, optional logging to S3, CloudTrail support, layer 4) vs ALB (single AZ, content-/host-/path-based routing, ECS dynamic port integration, HTTP/2, Web Sockets, HA {2 or more AZs}, WAF support, delete protection, X-Amzn-Trace-Id, layer 7), Proxy Protocol;
- EMR?;
- ENI?;
- FSMO role?;
- Glacier, cheap/slow data archival (3+ hours);
- HA:
- MySQL (async. replication);
- Oracle Database (DataGuard, RAC);
- SQL Server (AlwaysOn Availability Groups, clustering, mirroring);
- HTTP Live Streaming (HLS);
- HPC, Jumbo (Ethernet) Frames via Enhanced Networking (selected HVM instance types), PGs within an AZ;
- IAM:
- Cross-Account access, segregation of access for Dev., vs Test via pre-configured inline policy.; ie no need to remember a separate Account ID/username/password, can also be used to store/deploy SSL (in lieu of ACM);
- IDS/IPS, watch the AlertLogic video;
- Kinesis Data Streams, real-time data streaming (1-7 days);
- KMS, to generate signed certificates on demand for a requesting instance;
- Multicast?;
- NAT scaling
- OpsWorks, Chef 11+ deployments on AWS;
- Stacks, Layers (eg apps, caching, databases, load balancers), and Recipes;
- ELBs must be separately started up and attached initially, but subsequently are managed via OpsWorks;
- ELBs and SGs must be separately torn down after layer/stack deletion;
- Instances may be: 24/7(default), Time-based, and Load-based;
- Organizations:
- All Features vs Consolidated Billing: the former merely enables policy-based service controls for Accounts (eg deny EC2 in a bid to encourage Serverless computing);
- Consolidated Billing; ie a single bill for multiple Accounts, and with volume discounting too (eg EC2 RIs, S3):
- Alerts can still be individually configured at either level of the hierarchy;
- CloudTrail must be configured individually, logging to a Cross-Account S3 bucket, though;
- Promiscuous?;
- RDS:
- Multi-AZ (sync., durable) vs RRs (async., scalable);
- Multi-AZ tech.: AWS (Aurora/MariaDB/MySQL, Oracle Database, PostgresSQL), vs Microsoft (SQL Server mirroring);
- RRs in another Region OK, except Oracle Database/SQL Server;
- RRs can also be configured as Multi-AZ;
- RRs of RRs for MySQL only, and this will increase replica lag;
- Supports snapshotting to a different Region;
- Redshift (snapshot to S3, or copy to another Region), WLM;
- RIs:
- EC2, reserve within an AZ:
- On Demand: Unpredictable;
- Dedicated;
- Spot: Flexible provisioning, only if the bid price is met only;
- Reserved (Standard, Convertible, Scheduled): up to 75% discounts for 1-/3-year terms, for steady-state use (eg Production):
- May be split into multiple instances if the footprint remains the same;
- Restricted within the same family (eg T2) unless Convertible;
- Restricted for Linux only, excl. RHEL and SUSE;
- RDS, reserve within a Region, supports Multi-AZ and RRs (same Region only);
- Route 53;
- Routing symmetrically vs asymmetrically (ie round-trip data path);
- RTO vs RPO;
- S3 (11 9s durability), can be a VPC endpoint;
- Scale up (ie vertical), vs out (ie horizontal). Latter is preferred to minimise downtime;
- SES;
- SG, cannot setup explicit deny rules (NACLs can);
- Snowball/Snowmobile, or previously (data) Import/Export;
- SNS;
- SQS;
- Storage Gateway (on-prem {ESXi/Hyper-V} bandwidth-throttled, or as an EC2; also works with Direct Connect):
- File (NFS), up to 5 TB per file;
- Tape (iSCSI):
- Library (S3: instant);
- Shelf (Glacier: 1d);
- Volume (iSCSI):
- Cached (subset only, most frequently used, up to 32 volumes {32 TB ea}; ie 1 PB);
- Stored (full set, up to 32 volumes {16 TB ea}; ie 512 TB);
- SR-IOV?;
- STS, AD-based identity federation for 1-36 hour access (to some resource, eg S3) without having to create new IAM creds, LDAP authentication first (then STS), 4 fields (access key, secret access key, token, duration);
- SWF?;
- Tags are key/value pairs attached to resources, usable in Resource Groups;
- VM Import/Export;
- VPC tenancy (default vs dedicated) and its impact on EC2 instances;
- Route table (created by default), subnet to AZ (1:1), private vs public subnets, assign a public IP within a public subnet to make an instance internet-facing (behind an ELB also works), 5 reserved IPs per subnet (.0-.3, .255), CIDR block; “local” route within a VPC, IGW to VPC (1:1), route table (for n subnets), IGW/NAT target (for destination 0.0.0.0/0) route, SGs can span multiple subnets but not the other way around, NAT instance disable source/destination check, VPC peering: use private IPs to address instances within the same Region (50-125 VPCs), 1:1 relationship, private DNS names won’t resolve, routes/SGs/NACLs config required on both ends, multicast vs unicast?;
- WAF, managed layer 7 sandwich;