Evaluation Queue Rounds and Limits

Source material:

UI Design:
https://www.figma.com/file/E5zIAPsI2fagDc8btWGU06Ol/Challenge-Evaluation-Queue

Use case/desired behavior:
https://docs.google.com/document/d/19M9og2JW7CaG7YLNi8cCRhFuFwF-y6-nx4CJDAwr6qw/edit#heading=h.y0949lyt5na0

Motivations:

Existing Evaluations have a very inflexible way to define rounds:

{
	"name":"SubmissionQuota",
	"description":"Maximum submissions per team/participant per submission round.  If round information is omitted, then this indicates the overall submission limit per team/participant.",
	"properties":{
		"firstRoundStart": {
			"type": "string",
			"format": "date-time",
			"description": "The date/time at which the first round begins."
		},
		"roundDurationMillis":{
			"type": "integer",
			"description":"The duration of each round."
		},
		"numberOfRounds":{
			"type": "integer",
			"description":"The number of rounds, or null if there is no end."
		},
		"submissionLimit": {
			"type": "integer",
			"description":"the maximum number of submissions per team/participant per round."
		}
	}
}

The existing way to define Round limits assume that all rounds will last for the same duration and that later rounds will begin immediately as soon as the previous round ends. Users are also forced to perform unix timestamp calculations in order to set the round end datetime. There was also no fine grained submission limit control over Submission limits per user or participating team.

Proposed Changes

The goal is to provide users with more fine grained control over each Round in an Evaluation. This means:

Clearly defined start and end dates for each round.
- No more startDate + roundNumber * roundDruation math for the end date
Per-round submission limits. These limits should be changeable even during an ongoing round
- cumulative limit for the entire round
- per day - defined as every 24 hours from the start date
- per week - defined as every 7 days from the start date
- per month
  - monthly reset on every n-th day of the month. Use n==31, for end of month.
    - By default could be filled in with same day of month as the start of round
    - Requires adding a time zone field for the entire Evaluation queue since the day could be off by 1 depending on the time zone
  - OR monthly reset every 30 days.
Add/remove/modify rounds without affecting other rounds
- Change start date if round not yet started
- Change end date if round not yet ended
- Add/delete/change new rounds that start after the current time.
- Validation
  - Disallow intersecting time intervals between rounds
  - Optional: total limit > month limit > week limit > day limit
User’s Submissions will be automatically tagged with the index number of the current round
- Additional metadata column in Submission views
Allow Evaluation Queue Admins to schedule a Maintenance date range that disallows Submissions
- Independent of defined rounds - neither the ongoing rounds nor later rounds will have their start/end date modified as a result of maintenance

API Schema

Evaluation
Add timeZone, rounds. Deprecate quota

timeZone will use long names (e.g. “US/Pacific”) instead of 3-letter names (e.g. PDT, PST) to account for Daylight savings

{
	"description": "An Evaluation is the core object of the Evaluation API, used to support collaborative data analysis challenges in Synapse.",
	"name": "Evaluation",
	"properties": {
	
		"....currently existing fields...":{}


		"quota": {
			"type":"object",
			"description":"DEPRECATED. Maximum submissions per team/participant per submission round",
			"$ref":"org.sagebionetworks.evaluation.model.SubmissionQuota"
		},

		"timeZone": {
			"type": "string",
			"description": "The time zone to be used for all round configurations."
		},

		"rounds": {
			"type": "array",
			"items": {
				"$ref": "org.sagebionetworks.evaluation.model.SubmissionRound"
			}
		}
	}
}

EvaluationRound

Defines roundStart and roundEnd dates.

Question: should roundStart and roundEnd be purely integers? or allow ISO timestamps? Other fields in Evaluation(e.g. createdOn ) use “date-time” as type so we would maintain consistency, but it makes a rather odd paring to have timeZone='US/Pacific' and roundStart='2020-08-11T16:45:10−08:00'

id can used for updates if we store rounds in a separate table

submissionLimit is used to set submission limits

{
   "name":"EvaluationRound",
   "description":"Defines the duration of a round and sets limits for maximum submissions per round",
   "properties":{
      "id": {
         "type": "string",
         "description": "The id of the EvaluationRound"
      },
      "roundStart": {
         "type": "string",
         "format": "date-time",
         "description": "The date/time at which the first round begins.",
         "required": true
      },
      "roundEnd":{
         "type": "string",
         "format": "date-time",
         "description":"The date/time at which the round ends.",
         "required": true
      },
      "submissionLimit": {
         "type": {
            "$ref": "org.sagebionetworks.evaluation.model.EvaluationRoundLimit"
         },
         "description": "Optional. Sets limits for maximum submissions in this round."
      }
   }
}

EvaluationRoundLimit

Limits for day, week, month, and total

monthlySubmissionLimit is defined in a separate object

{
   "name":"EvaluationRoundLimit",
   "description": "Sets limits for maximum submissions in a EvaluationRound",
   "properties":{
      "totalSubmissionLimit": {
         "type": "integer",
         "description": "The maximum total number of submissions per team/participant for the entirety of the round."
      },
      "dailySubmissionLimit": {
         "type": "integer",
         "description": "The maximum total number of submissions per team/participant for each day of the round."
      },
      "weeklySubmissionLimit": {
         "type": "integer",
         "description": "The maximum total number of submissions per team/participant for every 7 days of the round."
      },
      "monthlySubmissionLimit": {
         "type": {
            "$ref": "org.sagebionetworks.evaluation.model.EvaluationRoundMonthlyLimit"
         },
         "description": "The maximum total number of submissions per team/participant per month. Unlike other limits, this can be configured to reset on the n-th day of every month.",

      }
   }
}

EvaluationRoundMonthlyLimit

{
   "properties": {
      "dayOfMonth": {
         "type": "integer",
         "description": "Day of the month on which this date this limit is reset."
      },
      "limit": {
         "type": "integer",
         "description": "The maximum total number of submissions per team/participant per month."
      }
   }
}

Example Evaluation

id will be back-end generated after it has been created

{
   "....currently existing fields...":{},
   
   
   "timeZone":"US/Pacific",
   
   "rounds":[
      {
         "id":"1",
         "roundStart":1231412213213123123,
         "roundEnd":"2020-08-11T16:45:10−08:00",
         "submissionLimit":{
            "totalSubmissionLimit":20,
            "dailySubmissionLimit":2,
            "weeklySubmissionLimit":8,
            "monthlySubmissionLimit":{
               "dayOfMonth":28,
               "limit":10
            }
         }
      },
      
      
      {
         "id":"2",
         "roundStart":1231412213213123123,
         "roundEnd":"2020-08-11T16:45:10−08:00",
         "submissionLimit":{
            "totalSubmissionLimit":40,
            "dailySubmissionLimit":8,
            "weeklySubmissionLimit":12,
            "monthlySubmissionLimit":{
               "dayOfMonth":31,
               "limit":20
            }
         }
      }
      
      
   ]
}

EvaluationMaintenance

disallows all submissions
only 1 allowed per queue

{
   "properties": {
      "start": {
         "type": "string",
         "format": "date-time",
         "description": "The date/time at which the maintenance begins.",
         "required": true
      },
      "end":{
         "type": "string",
         "format": "date-time",
         "description":"The date/time at which the maintenance ends.",
         "required": true
      },
   }
}

API Endpoints

Endpoint	Request body	Return Body	Description
GET /evaluation/{evalId}	None	Evaluation	EXISTING endpoint to get evaluation
POST /evaluation	Evaluation	Evaluation	EXISTING endpoint to create an evaluation
PUT /evaluation/{evalId}	Evaluation	Evaluation	EXISTING endpoint to update evaluation
POST /evaluation/{evalId}/maintainance	EvaluationMaintenance	None or EvaluationMaintenance	Schedule a maintainance. Additional calls will override the existing maintenance window
DELETE /evaluation/{evalId}/maintainance	None	None	Delete the existing maintenance window, if any
GET /evaluation/{evalId}/maintainance	None	EvaluationMaintenance	Get info about the maintenance window. (used for UI display)

API/Database Transition

A SubmissionQuota can easily be mapped into into a list of SubmissionRounds using math, but the reverse is not true.

For the time being, I’m considering making the old deprecated quota field remain stored in the database, but enforce a rule such that user must choose between quota and rounds when configuring a Evaluation. This allows time for clients to implement support for the new rounds field.
Once we fully decide to remove SubmissionQuota, we can perform the conversion and store quotas asSubmissionRound.

Database Tables

2 Options:

Store rounds List as JSON in a single database column.
- This would require searching for the correct round based on start/end dates in memory
- Evaluations object can be cleanly retrieved using query on single table
Separate Table For Rounds (primary key is evaluationId , ID) index on roundStart and roundEnd
- Allows us to pull out one specific limit
- Separate queries
  - query to retrieve list of all rounds for that evaluation,
    - On REST API GET, we care about all rounds
  - query to retrieve list single evaluation id,
    - On challenge submissions during the submission limit enforcement, we only care about the specific round whose start/end interval encapsulate the current time
  - query to pull from the Evaluations table
- Limits are enforced in Java code so we can store the them as JSON
- Question: should ID be created by the id-generator? or just index in rounds list?
  - generated unique ID makes it easy to perform updates on evaluations
    - with current API setup a list of round are submitted, making this pointless
  - index in round list as ID
    - this allows easier tagging of submission
    - updating rounds involves row deletion and addition if added
      - hash metdata to avoid pointless delete/add?
  - If we use roundStart, roundEnd to identify the round config.
    - any change to start,roundend would mean an insertion/deletion

evaluationId(BIGINT)(Foreign Key to Evaluations Table)	ID(BIGINT)	roundStart(TIMESTAMP)	roundEnd(TIMESTAMP)	Limits(JSON)
123	1	`2020-01-01 04:40:10`	`2020-01-01 04:40:10`	{ "id":"2", "roundStart":1231412213213123123, "roundEnd":"2020-08-11T16:45:10−08:00", "submissionLimit":{ "totalSubmissionLimit":40, "dailySubmissionLimit":8, "weeklySubmissionLimit":12, "monthlySubmissionLimit":{ "dayOfMonth":31, "limit":20 } } }
123	2	`2020-01-01 04:40:10`	`2020-01-01 04:40:10`	…
456	1	`2020-01-01 04:40:10`	`2020-01-01 04:40:10`	…
789	1	`2020-01-01 04:40:10`	`2020-01-01 04:40:10`	…

Evaluation_maintenance table

Records maintenance windows for each evaluation.

evaluationId(BIGINT)(Foreign Key to Evaluations Table)	startDate(TIMESTAMP)	endDate(TIMESTAMP)
123	`2020-01-01 04:40:10`	`2020-01-01 04:40:10`
456	`2020-01-01 04:40:10`	`2020-01-01 04:40:10`

JDOSUBMISSION table

Add EVALUATION_ROUND, which is the round number during which it was submitted. This will be replicated into Submission Views.

Existing table:

ID BIGINT NOT NULL,
NAME varchar(256) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci,
EVALUATION_ID BIGINT NOT NULL,
USER_ID BIGINT NOT NULL,
SUBMITTER_ALIAS varchar(256) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci,
ENTITY_ID BIGINT NOT NULL,
ENTITY_BUNDLE mediumblob,
ENTITY_VERSION BIGINT NOT NULL,
CREATED_ON BIGINT NOT NULL,
TEAM_ID BIGINT,
DOCKER_REPO_NAME varchar(400) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci,
DOCKER_DIGEST varchar(200) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci,