Add txObservers and make use of TXN Stats Collector object

[nklincoln]

BIG change here brings in transaction observer modules, and with it some other somewhat significant breaking changes that will need equally significant documentation changes:

Transaction observers are specified within the benchmark config file to be used by workers. They may use:

• logging (used to log transactions)
• prometheus-push (to push updates via a push gateway for scrape by prometheus)

An internal observer is always used, which is used to:

• send updates to the manager process that might have enabled the default observer
• send completion to the round orchestrator to use in building reports (there is only one way to build txStatistics report information, as the prometheus method is now invalid)
• txUpdate interval is configured via internal property caliper-progress-reporting-interval (in ms)
• txUpdate is disabled via internal property caliper-progress-reporting-enabled (true/false)

The local-observer is renamed to default-observer.

MessageHandler is renamed WorkerMessageHandler

All transaction statistics are collected, passed and manipulated via the TransactionStatisticsCollector class.

Workers no longer build an array of transactions to Promise.all in order to assure completion, we monitor for completion by taking advantage of aggregated results in the TransactionStatisticsCollector instance.

Rate controllers use the worker TransactionStatisticsCollector instance to determine required sleep times for rate control purposes

FixedBacklog controller renamed FixedLoad, since it aims to maintain a fixed txn load on the SUT

Updated generators to set different default values for txDuration and txNumber tests, as well as reduce the number of default workers (mainly to help the build).

Resource manager specification has been changed to mirror the specification of txStatsCollectors. This unifies the specification within the configuration file. A complete specification for monitors (transaction and resource, as per the integration tests) is:

monitors:
    transaction:
    - module: logging
      options:
        loggerModuleName: txinfo
        messageLevel: info
    - module: prometheus-push
      options:
        interval: 5
        push_url: "http://localhost:9091"
    resource:
    - module: prometheus
      options:
        interval: 5
        url: "http://localhost:9090"
        metrics:
            ignore: [prometheus, pushGateway, cadvisor, grafana, node-exporter]
            include:
                Endorse Time (s):
                    query: rate(endorser_propsal_duration_sum{chaincode="marbles:v0"}[1m])/rate(endorser_propsal_duration_count{chaincode="marbles:v0"}[1m])
                    step: 1
                    label: instance
                    statistic: avg
                Max Memory (MB):
                    query: sum(container_memory_rss{name=~".+"}) by (name)
                    step: 10
                    label: name
                    statistic: max
                    multiplier: 0.000001
        charting:
            polar:
                metrics: [Max Memory (MB)]
            bar:
                metrics: [all]
    - module: process
      options:
        interval: 3
        processes: [{ command: 'node', arguments: 'caliper.js', multiOutput: 'avg' }]
    - module: docker
      options:
        interval: 4
        containers: ['peer0.org1.example.com', 'peer0.org2.example.com', 'orderer0.example.com', 'orderer1.example.com']

closes #884
closes #826
closes #645

Signed-off-by: [email protected] [email protected]

[aklenik]

Amazing PR, long needed!!
I had some minor comments, but only the MessengerInterface ones are in the must-fix category.

[aklenik]

The enabled flag could be derived from interval !== 0. But I'm also fine with this explicit flag.

[lgtm-com]

This pull request introduces 1 alert when merging e9aa4fc into c84d59f - view on LGTM.com

new alerts:

• 1 for Unused variable, import, function or class