CloudServices/DataPipeline
Jump to navigation
Jump to search
Overview
The cloud services data pipeline ingests data for analysis, monitoring and reporting. The pipeline is currently used for processing cloud services server logs. We're in the process of improving it to support desktop and device telemetry data. The data pipeline team also works on Heka (a major component of the pipeline implementation), custom dashboards for cloud services projects, and the Telemetry server.
Resources
- IRC channel: #datapipeline
- Standup meeting: https://etherpad.mozilla.org/data-pipeline-meeting-notes
- Cross team coordination meeting: https://etherpad.mozilla.org/data-pipeline-coordination
- Pipeline technical proposal
- Reporting and monitoring overview
- Cloud Services Data Projects
- List of Data Sources
Pipeline Milestones
- Q4 2014: Telemetry data running through pipeline
- Server stack deploy in github ("opsified")
- Re-implement monitoring dashboards
- Q1 2015: Launch pipeline prototype
- Architecture decisions completed; production stack up and running
- Business Intelligence/Data Warehouse proof of concept implemented
- Ingestion process completed for FHR+telemetry (start collecting on 2015-02-23)
- Backprocessing from pipeline datastore implemented
- Pipeline runs in parallel to existing infrastructure; not yet source of truth
- Q2 2015: Pipeline officially supports business use cases
- Complete set of use cases tbd (most likely primarily FHR+telemetry use cases)
- Complete set of monitoring and reporting outputs tbd: dashboards, data warehouse, monitoring, self-service access to data
- FHR+telemetry hits full release 2015-05-19, handle full production load
- Q3 2015: Fill out monitoring and reporting capabilities; add sources and use cases
Related Dates and Schedules
- FHR+Telemetry client work
- Current plan: FF39 Nightly and uplifted to FF38. May not hit this schedule, but the pipeline needs to be ready
- 2015-02-23 Nightly
- 2015-05-19 Release
Work Queue
To Do/Doing categories need to be logged in bugzilla.
Risks and Open Questions
- Old-FHR data through pipeline? Yes/No: [telliot]
- Deletes & legal policy [telliot, mreid to provide cost estimate]
Needs more discussion
- Maintaining a sample data set for faster queries
- Implement a specific flag to determine if data gets warehoused or not
- Integrate Roberto’s spark data flow into new DWH
- Implies a similar db-backed table of DWH filenames for filtering (don’t want to list S3 every time - too slow)
- Elasticsearch (Kibana) output filter
- Complete list of outputs (and filters and any other support)
- Build a shim for debugging CEPs with local data
- Store the “raw raw” data for some period to ensure we’re safe if our code and/or CEP code is badly broken. Can’t just lose data.
- Tee off to short-lived S3 before it goes through the main pipeline?
- BI query example that cross references data sources
- example: does fxa/sync increase browser usage?
To Do
- Q4 telemetry: (re) implement telemetry monitoring dashboards [?]
- Q1 BI: define schema for data warehouse (talk to jjensen) [kparlante]
- should use multiple data sources
- Q1 BI: write filter for data warehouse [trink]
- Q1 BI: signal & schedule loading of data warehouse [mreid]
- Q1 BI: redshift output [trink]
- Q1 BI: setup domo and/or tableau to look at mysql or csv or whatever is easy [?]
- Q1: Data format spec [kparlante, trink]
- JSON schema, specifically for FHR+telemetry, also anticipate other sources
- Q1: implement best guess at per user sampling [trink]
- follow up with saptarshi for more complex algorithm
Doing
- Opsify stack [whd]
- Q4 telemetry: Send telemetry data through the pipeline [mreid]
- Q4 telemetry: Larger payloads (32MB) for telemetry [trink]
- risk mitigation: Estimate cost of “full scan” DWH query [mreid]
- risk mitigation: Estimate cost of single DWH delete [mreid]
Done
- Parallelize sandbox filters (eg FHRSearch) [trink]
- Enable Lua JIT [trink]