Download as PDF, PPTX
Uploaded bySalesforce Developers
Salesforce API Series: Fast Parallel Data Loading with the Bulk API Webinar
This document summarizes a Salesforce webinar about loading data in parallel using the Bulk API. It discusses how parallel processing can significantly increase data load throughput compared to serial loads. However, locks and other inhibitors can prevent optimal parallelism. The webinar demonstrates approaches to identify and manage locks, such as modifying the data schema, ordering the load file, and using a controlled parallel approach. Managing locks is key to achieving high parallelism and throughput during large data loads.
In this document
Powered by AI
Introduction to Salesforce API Series focusing on Bulk API for fast data loading and Safe Harbor statement.
Discussion on the speed of data loading into Salesforce and inquiry into how many records can be loaded in an hour.
Graphs illustrating data load throughput (up to 25M records/hour) and explanations of parallel vs. serial processing.
Hands-on activities contrasting serial and parallel data loads, measuring throughput, and identifying inhibitors.
Overview of the Salesforce Bulk API, its asynchronous data loading capabilities, and job configuration for serial loads.
Summary of serial load results showing 1 million records loaded in 52 minutes with specific throughput metrics.
Insights into expected plans for parallel loads and the impact of lock exceptions on throughput and performance.
Methods to optimize parallel loads by eliminating locks, achieving high throughput (over 320,000 records/min) with results.
Discussion on managing locks during parallel loads, with samples showing successful load processes with minimal failures.
Explanation of controlled feed loading methods emphasizing explicit throttling, prioritizing asynchronous processes.
Recap of key concepts on managing locks in parallel loads followed by a question and answer session with the speakers.
More Related Content
Similar to Salesforce API Series: Fast Parallel Data Loading with the Bulk API Webinar
Salesforce API Series: Fast Parallel Data Loading with the Bulk API Webinar
- 1. Salesforce API Series FastParallel Data Loading with the Bulk API February 26, 2014
- 2. Safe Harbor Safe harborstatement under the Private Securities Litigation Reform Act of 1995: This presentation may contain forward-looking statements that involve risks, uncertainties, and assumptions. If any such uncertainties materialize or if any of the assumptions proves incorrect, the results of salesforce.com, inc. could differ materially from the results expressed or implied by the forward-looking statements we make. All statements other than statements of historical fact could be deemed forward-looking, including any projections of product or service availability, subscriber growth, earnings, revenues, or other financial items and any statements regarding strategies or plans of management for future operations, statements of belief, any statements concerning new, planned, or upgraded services or technology developments and customer contracts or use of our services. The risks and uncertainties referred to above include – but are not limited to – risks associated with developing and delivering new functionality for our service, new products and services, our new business model, our past operating losses, possible fluctuations in our operating results and rate of growth, interruptions or delays in our Web hosting, breach of our security measures, the outcome of intellectual property and other litigation, risks associated with possible mergers and acquisitions, the immature market in which we operate, our relatively limited operating history, our ability to expand, retain, and motivate our employees and manage our growth, new releases of our service and successful customer deployment, our limited history reselling nonsalesforce.com products, and utilization and selling to larger enterprise customers. Further information on potential factors that could affect the financial results of salesforce.com, inc. is included in our annual report on Form 10-Q for the most recent fiscal quarter ended July 31, 2012. This documents and others containing important disclosures are available on the SEC Filings section of the Investor Information section of our Web site. Any unreleased services or features referenced in this or other presentations, press releases or public statements are not currently available and may not be delivered on time or at all. Customers who purchase our services should make the purchase decisions based upon features that are currently available. Salesforce.com, inc. assumes no obligation and does not intend to update these forward-looking statements. #forcewebinar
- 3. Speakers Steve Bobrowski Architect Evangelist @sbob909 #forcewebinar SeanRegan Architect Evangelist @sfdcsregan
- 4. Follow Developer Forcefor the Latest News @forcedotcom / #forcewebinar Developer Force – Force.com Community +Developer Force – Force.com Community Developer Force Developer Force Group #forcewebinar
- 5. How fast canyou load data into Salesforce?
- 6. How many recordscan you load into Salesforce in 1 hour?
- 7.
- 8.
- 9. A parallel processinganalogy: digging a ditch #forcewebinar
- 10.
- 11.
- 12. The number ofprocesses or threads associated with an operation.
- 13. Optimal parallel processing 5Mrecords Parallel 5M records 5M records 5M records Serial 20M records Time #forcewebinar
- 14. Sub-optimal parallel processing 5Mrecords Parallel 5M records 5M records 5M records Serial 20M records Time #forcewebinar
- 15. Locks, exceptions, triggers,relationships, … 5M records Parallel 5M records 5M records 5M records Serial 20M records Time #forcewebinar Throughput inhibitors
- 16. Data load casestudies § Get hands on with the Salesforce Bulk API § Contrast serial data loads vs. parallel data loads § Measure degrees of parallelism and throughput § Identify and avoid throughput inhibitors § Achieve maximum throughput #forcewebinar
- 17.
- 18. Salesforce Bulk API § Asynchronous data loading § Optimized for large data sets § REST API § Powers many tools § Use to build custom tools with any programming language (Java, etc.) #forcewebinar
- 19.
- 20. Bulk API Loadsthat … ealize, nvestigate, and lan
- 21.
- 22.
- 23. Serial load: Expectedplan Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread • • • • Time #forcewebinar One job 100 batches 10,000 records/batch 1M total records
- 24. Serial load: Jobconfiguration #forcewebinar
- 25. Serial load: Batchcreation #forcewebinar
- 26. Serial load: Batchrun #forcewebinar
- 27.
- 28. Serial load summary ConcurrencyMode Records Loaded Records Failed Serial 1 million 0 Run Time 52 minutes Work Completed 48 minutes Throughput Degree of Parallelism Key Problem Solution 19,500 records per minute 0.94 Degree of parallelism explicitly limited to ~1. Explore parallel load for increased throughput. #forcewebinar
- 29. Throughput Records/Min Parallelism vs.Throughput of a Single Job 350000 Serial Run • Low degree of parallelism 300000 250000 200000 150000 100000 50000 Serial 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Degree of Parallelism #forcewebinar
- 30.
- 31. Parallel load: Expectedplan Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread Thread • • • • One job 100 batches 10,000 records/batch 1M total records Time #forcewebinar
- 32. Parallel load: Jobconfiguration #forcewebinar
- 33. Things to watchfor § Locks can significantly affect parallel loads – Wasted processing capacity – Reduced throughput – Failures § Retry logic is not all its cracked up to be #forcewebinar
- 34.
- 35. Parallel load 1summary Concurrency Mode Records Loaded Records Failed Parallel 125,000 875,000 Run Time 10 minutes Work Completed 2 hours and 30 minutes Throughput Degree of Parallelism Key Problem Solution 20,000 records per minute 15.79 Lock Exceptions. Server worked significantly harder but no increase in throughput. Run the load in serial mode or manage locks. #forcewebinar
- 36. Throughput Records/Min Parallelism vs.throughput of a single job 350000 Parallel Run 1 • High degree of parallelism • Low throughput due to locks 300000 250000 200000 150000 100000 50000 Serial Parallel 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Degree of Parallelism #forcewebinar
- 37. Time to optimize § § Let’smake your data load ealize – Locks inhibit parallelism and throughput § nvestigate – What is causing the locks § lan – Manage the locks #forcewebinar
- 38. Demo Parallel load 2 EliminateLocks by Modifying Schema
- 39. Parallel load: Sampleresults Concurrency Mode Records Loaded Records Failed Parallel 1 million 0 Run Time 3 minutes and 30 seconds Work Completed 1 hour Throughput Degree of Parallelism Key Problem Solution 320,000 records per minute 19 None n/a #forcewebinar
- 40. Throughput Records/Min Parallelism vs.throughput of a single job 350000 Parallel 2 Parallel Run 2 • High degree of parallelism • High throughput 300000 250000 200000 150000 100000 50000 Serial Parallel 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Degree of Parallelism #forcewebinar
- 41. Locks can bemanaged by § Elimination § Ordering load file #forcewebinar
- 42. Demo Parallel load 3 AvoidLocks with Ordered Data
- 43. Managing locks …a discussion while we load § Master-detail relationships § Lookup relationships § Roll-up summary fields § Triggers § Workflow rules § Group membership locks* #forcewebinar
- 44. Parallel load: Sampleresults Concurrency Mode Records Loaded Records Failed Parallel 1 million 0 Run Time 4 minutes Work Completed 1 hour Throughput Degree of Parallelism Key Problem Solution 250,000 records per minute 16.5 Minimal overhead due to locks Remove all unnecessary locks #forcewebinar
- 45. Throughput Records/Min Parallelism vs.throughput of a single job 350000 Parallel Run 3 • High degree of parallelism • High throughput 300000 250000 Parallel 2 Parallel 3 200000 150000 100000 50000 Serial Parallel 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Degree of Parallelism #forcewebinar
- 46.
- 47. Controlled feed loadmethodology § Explicit throttling on parallelism and throughput – Parallel extraction and loading – Prioritization of asynchronous processing capacity § Manage inhibitors in complex jobs – Data Skews – Multiple Locks #forcewebinar
- 48. Throughput Records/Min Parallelism vs.throughput of a single job 350000 Parallel 2 Controlled Feed Run • Reduced parallelism • Expected throughput 300000 250000 Parallel 3 200000 150000 100000 Controlled Feed 50000 Serial Parallel 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Degree of Parallelism #forcewebinar
- 49. Related wiki articleand Architect Core Resources #forcewebinar
- 50. Recap § § Make your paralleldata loads ealize – Locks inhibit parallelism and throughput § nvestigate – What is causing the locks § lan – Manage the locks #forcewebinar
- 51. Q&A Steve Bobrowski Architect Evangelist @sbob909 #forcewebinar SeanRegan Architect Evangelist @sfdcsregan