When Web Services Go Bad
Download as PPT, PDF1 like913 views
The document discusses some of the challenges of developing and deploying web services at scale, including: - Meeting service level agreements for high availability and performance. - Choosing appropriate technologies and architectures that can scale to support large volumes of traffic and data. - Ensuring services are robust, reliable and secure through practices like rigorous testing, monitoring, and automated deployment. - Fostering collaboration between development and operations teams to address deployment issues as they arise.
Ad
More Related Content
What's hot (20)
Viewers also liked (19)
Ad
Similar to When Web Services Go Bad (20)
Ad
More from Steve Loughran (20)
Recently uploaded (20)
When Web Services Go Bad
- 1. when web services go bad steve loughran hp laboratories March 2002 notes from the field
- 2. global distribution global load expectations of “web speed” development http at the bottom integration with remote callers service level agreements covering QoS a new set of a problems! why is it so hard? web service development =
- 3. image storage and SVG rendering service XML-RPC POST SVG in payload render to JPEG return URL user fetch or print on press +1TB image store for customer photos
- 4. constraints SLA: high availability render times: 2s proof render 15s production basic asset store done: -SQL server IIS/ASP COM/MTS->Win2K/COM+ “real soon now” XML-RPC agreed on RPC mechanism wildly optimistic timescales
- 5. SVG into XML-RPC <?xml version="1.0"?><methodCall version="1.0"> <methodName>render_svg</methodName> <params> <param><value><string> 110 </string></value></param> <param><value> <string> PD94bWwgdmVyc2lvbj0iMS4wIiBzdGFuZGFsb25lPSJubyIgPz4NCjwhRE9DVFlQRSBzdmcgUFVC TElDICItLy9XM0MvL0RURCBTVkcgMjAwMDExMDIvL0VOIg0KICJodHRwOi8vd3d3LnczLm9yZy9U Ui8yMDAwL0NSLVNWRy0yMDAwMTEwMi9EVEQvc3ZnLTIwMDAxMTAyLmR0ZCI+DQo8c3ZnIHdpZHRo PSIxMDI0cHQiIGhlaWdodD0iMTAyNHB0Ij4NCiA8ZyBpZD0icGljdHVyZSIgPjxpbWFnZSB3aWR0 aD0iMTAyNHB0IiBoZWlnaHQ9Ijc2OHB0Ig0KICAgeGxpbms6aHJlZj0iaHR0cDovL3N0ZWx2aW86 ODA4MC9zdW5zZXQuanBnIi8+PC9nPg0KPHRleHQgeD0iMjlwdCIgeT0iODVwdCINCiBzdHlsZT0i Zm9udC1mYW1pbHk6SGVsdmV0aWNhO2ZvbnQtc2l6ZTozNnB0O2ZpbGw6I2QwZDBkMDsiDQo+a2Vp dGggaXMgKG1heWJlKSBzZXh5PC90ZXh0Pg0KPC9zdmc+DQoNCg0K</ string></value> </param> <param><value><int> 72 </int></value></param> </params></methodCall> <?xml version="1.0" standalone="no" ?> <!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 20001102//EN" "http://www.w3.org/TR/2000/CR-SVG-20001102/DTD/svg-20001102.dtd"> <svg width="1024pt" height="768pt"> <g id="picture" ><image width=“1024pt" height=“768pt" xlink:href="http://stelvio:8080/sunset.jpg "/></g> <text x="29pt" y="85pt" style="font-family:Helvetica;font-size:36pt;fill:#d0d0d0;“> Having an excellent holiday! </text> </svg>
- 6. Response <?xml version="1.0"?> <methodResponse> <params> <param><value><string>2453</string></value></param> <param><value><string> http://stelvio:8080/hpsvgservlet/fetch?ref=_2_ec5132765b_4553.jpg-110 </string></value> </param> </params> </methodResponse>
- 7. what we ended up with ArrowPoint Apache Apache Bluestone x4 Bluestone x4 IIS/ASP IIS/ASP IIS/ASP PDC + DNS file store SQL server SQL server remote renderer store
- 8. what worked? Web Service Model! XML based RPC HTTP for transport SVG for request ANT build & deploy JUnit for unit tests incremental development
- 9. (nearly) no more WORKSFORME all server-side problems could be replicated log all failures server side for easier post mortem SEH catching of all win32 renderer errors phase II: SEH errors => auto email to support alias just integration… networking, client code issues (proxies, wrong URL, authentication)
- 10. ANT + JUnit fully automated build process integrated JUnit testing automated deployment to local server stacks
- 11. security cookie authentication in common domain encrypt user, session in cookie restricted ports on beta site admin pages by IP addr sanitize incoming SVG XML: -catch file: references -downgrade http: access -test XML &includes;
- 12. configuration ASP: config in source: brittle, scaling issues Java: per cluster config files in the WAR +kept under SCM; cleaner -delays to change, scaling Use a database ? Use a directory service?
- 13. what didn’t work: turning a web site into a web service
- 14. what went wrong FedEx cabling raid controller unreliable switch router config forgotten passwords IP address issues accidental deletion of 8GB test data (!) COM+ authentication DNS JVM ‘lockup’ time differences between servers cluster race conditions boot race conditions MP thread safety resource leakage errors in JSP/ASP show up after deployment
- 15. operations paranoia R&D not allowed near production boxes response to any security issue is “no live” when things don’t work, then they call us escalated minor security niggles into blocking issues effect #1: threat to weekends effect #2: threat to schedule effect #3: we stopped telling them of “issues”
- 16. error messages java.io.NoRoute ToHostException if you don’t have Java/C# engineers on the ops team, you get called in for every message fix #1: “defect” tracking of operations issues, from early days of the app fix #2: always identify the errant system in fault codes
- 17. HTTP Content-length is all you get how do you propagate a transient failure E_TRANSIENT_FAILURE =>retry with exponential backoff -how you really test this? -what about integrity?
- 18. firefighting it takes 10 minutes to deploy it takes 30s to report a defect therefore, it must take 11 minutes to fix a bug and deploy the update … stay in control by never updating public servers more frequently than nightly
- 19. complexity is the enemy availability is the primary casualty of complexity availability A(X) ≝ P(X) is working 0 ≤ A(X) ≤ 1 service S depends on services s 1..n A(S) = A(node)*A(s 1 )* A(s 2 )…A(s n ) redundancy: A(n 1 +n 2 )=A(n 1 )+A(n 2 )-A(n 1 )A(n 2 ) cost of redundancy depends on scalability of service: O(1), O(n), O(n 2 ), O(2 n ) …
- 20. a modest proposal: deployment-centric software processes
- 21. operations use cases/ XP stories update live server add new fonts bind to new database change account/passwords backup/restore system partition cluster multi-home the server diagnose intermittent failure these need support in software and/or process
- 22. deployment and operations test cases probe for needed exes, COM objects validate remote sites visible check configuration treat all deployment issues as defects to track and to have test cases run the test cases at install time, and from the load balancing router
- 23. operations issues are defects treat all deployment issues as defects to track don’t just fix it once, at it will crop up again. you need regression tests you need a repository of defects for easy searching. or they will phone you at 3am
- 24. deploy early deploy often involve operations in system design give them regular builds to deploy during early development use a tool like CruiseControl for continual integration and local deployment provide 5x6 support at this stage maintain slow bug fix times
- 25. instrument with JMX (Java) WMI (.NET) a public class ServiceManager implements ServiceManagerMBean { protected Service _owner; protected int _serviceTransactionCount; protected int _serviceUseCount; protected double _sellThroughDollars; public synchronized void bookSales(int count, double dollars) { _sellThroughDollars+=dollars; _serviceUseCount+=count; _serviceTransactionCount++; } public Double getSellThroughDollars() { return new Double(_sellThroughDollars); }
- 26. current work SmartFrog Distributed deployment framework Configuration is Deployment declare desired state of machines Runtime handles it www.smartfrog.org
- 27. How to host big applications across distributed resources Automatically / Repeatably Dynamically Correctly Securely How to manage them from installation to removal How to make grid fabrics useful for classic server-side apps HPLabs research
- 28. SmartFrog Deployment Engine SmartFrog Node SmartFrog Components Description / Code Repositories RMI RMI Deploy Descriptions SmartFrog Daemon SmartFrog Node SmartFrog Components SmartFrog Daemon SmartFrog Node SmartFrog Components SmartFrog Daemon RMI / (SOAP) declare desired state of the system runtime(s) instantiate components components configure, start and stop the apps
- 29. summary web services are more complex than web sites or intranet applications to deploy developers: deployment is (nearly) everything operations: developers are not your enemy management: recognize the new problems; address them strive for simplicity
Editor's Notes
- #2: 06/10/09 This is meant to conjure up the vision of some late night cable TV show “we take you behind the scenes of colocation sites, finding the worst web services in existence”, interviewing the people using them, managing them, integrating them, before finally catching up with the developer team on their doorsteps, asking them “why did you produce such a nightmare?” If such a show existed, would you be on it? I am going to tell you how to avoid that, without getting the government to give you a new identity under the Developer Relocation Program. Would I be on it? I would have liked to have been on it before the project was over. This is a photo of me 6000 foot up one of the cascade peaks on a technical spring mountaineering weekend, and I was getting voicemail about config problems. It would have been nice to have had a new identity then. As it is I had feign cellphone coverage failure