Skip to content

Files

Latest commit

diegojancicdiegojancic
added slim_handler comparison
Dec 19, 2018
ce8e1f9 · Dec 19, 2018

History

History
43 lines (28 loc) · 2.39 KB

zappa_slim_handler.md

File metadata and controls

43 lines (28 loc) · 2.39 KB

Zappa slim_handler Performance Results

Introduction

Comparison of 2 similar functions. Both function are developed using Zappa and Flask, and include an extra binary file of 46.1mb making the resulting Lambda package of 49.6mb, which is within the 50mb limit. One Lambda is deployed with slim_handler=True and the other one with slim_handler=False.

Packages Sizes

slim_handler=True:

  1. Lambda function size: 10.64mb
  2. Package size: 53.86mb

slim_handler=False:

  1. Package size: 49.6mb

Execution Times and Memory Usage

FRAMEWORK BOOTSTRAP RESULTS
Function Size(MB) VPC D.Mean Var Billed Mean Var Mem Mean Var Samples
zappa-slim 10.64 No 5882.98 322278 5936.73 321122 166.857 14.2083 49
zappa-noslim 49.59 No 835.252 18637.4 884 18922.4 46.98 9.85673 50
WARM START RESULTS
Function Size(MB) VPC D.Mean Var Billed Mean Var Mem Mean Var Samples
zappa-slim 10.64 No 3.85507 26.362 100 0 167.736 15.1235 4918
zappa-noslim 49.59 No 4.1465 28.8056 100 0 49.4513 9.52658 5010

Conclusions

  1. Looking at both packages, the one with the slim_handler, includes botocore, making the packages bigger.
  2. Using slim_handler increases the memory usage significantly (+118mb). Based on AWS information, the usage of additional memory leads to the Lambda functions to be recycled more often.
  3. When using slim_handler, the cold starts (without including the function execution time) will be up to 660ms faster as the package is smaller. (see Results BY FUNCTION SIZE)
  4. There's no difference in time execution for warm lambdas (even though the extra memory usage could make things slower if scarce).
  5. For cold-starts, the slim_start adds an extra of at least 5 seconds in execution time. If the associated package is larger than 50mb, the extra time could be even larger.