https://www.growthbook.io/ logo
#announcements
Title
# announcements
h

hundreds-belgium-41942

07/13/2023, 10:33 AM
Hi, I'm getting this error whenever i try to run the query. This has been happening for me for 2 weeks now.
w

white-fireman-22476

07/13/2023, 1:15 PM
Probably it is some json field in the result set that is undefined unexpectedly. You can see what query is being run by clicking on the show sql link and then run it against your database, to see if something is unexpectedly empty.
h

hundreds-belgium-41942

07/14/2023, 3:51 AM
@cuddly-fireman-12522
c

cuddly-fireman-12522

07/20/2023, 8:42 AM
@white-fireman-22476 This is the error we are getting from container
Copy code
{
  "level": 50,
  "time": 1688636369686,
  "pid": 59,
  "hostname": "fa961f2113a5",
  "err": {
    "type": "SyntaxError",
    "message": "Unexpected token u in JSON at position 0",
    "stack": "SyntaxError: Unexpected token u in JSON at position 0\n    at JSON.parse (<anonymous>)\n    at /usr/local/src/app/packages/back-end/dist/services/stats.js:99:27\n    at Generator.next (<anonymous>)\n    at fulfilled (/usr/local/src/app/packages/back-end/dist/services/stats.js:5:58)\n    at processTicksAndRejections (node:internal/process/task_queues:96:5)"
  },
  "msg": "Failed to run stats model: null"
}
w

white-fireman-22476

07/20/2023, 9:01 AM
Hi. That is helpful. I imagine you might get this error if the python code errored. Since you are running it in the container, I would assume that python is set up correctly and would be the same for you as everyone else. The code that python runs though can vary based upon your variation names, among other things, but the names seems like the most likely culprit to me. Are you using any special characters in the names of your variations?
c

cuddly-fireman-12522

07/20/2023, 9:05 AM
Hi James, we don’t use any special characters. Anyway, this was working fine before, but occasionally this error came up. But now every time we try to update data in experiments we get this error
w

white-fireman-22476

07/20/2023, 9:19 AM
hrmm, well there must be something that is making you unique, and Growthbook is not handling it correctly. Could you maybe add a debug line to console.log() everything from line 52 to line 124 in packages/back-end/src/services/stats.ts and let me know what it is? Hopefully we can solve the bug for you and everyone else who might face your situation.
c

cuddly-fireman-12522

07/20/2023, 9:20 AM
I will check and let you know James. Thanks!
w

white-fireman-22476

07/20/2023, 9:21 AM
Thank you! Based upon what you have shared at this point, its probably a bug on our side.
Copy code
import { promisify } from "util";
import { PythonShell } from "python-shell";
import {
  DEFAULT_SEQUENTIAL_TESTING_TUNING_PARAMETER,
  DEFAULT_STATS_ENGINE,
} from "shared/constants";
import { MetricInterface } from "../../types/metric";
import { ExperimentMetricAnalysis, StatsEngine } from "../../types/stats";
import {
  ExperimentMetricQueryResponse,
  ExperimentResults,
} from "../types/Integration";
import {
  ExperimentReportResultDimension,
  ExperimentReportResults,
  ExperimentReportVariation,
} from "../../types/report";
import { promiseAllChunks } from "../util/promise";
import { checkSrm } from "../util/stats";
import { logger } from "../util/logger";
import {
  ExperimentSnapshotAnalysisSettings,
  ExperimentSnapshotSettings,
} from "../../types/experiment-snapshot";
import { QueryMap } from "../queryRunners/QueryRunner";

export const MAX_DIMENSIONS = 20;

export async function analyzeExperimentMetric(
  variations: ExperimentReportVariation[],
  metric: MetricInterface,
  rows: ExperimentMetricQueryResponse,
  dimension: string | null = null,
  statsEngine: StatsEngine = DEFAULT_STATS_ENGINE,
  sequentialTestingEnabled: boolean = false,
  sequentialTestingTuningParameter: number = DEFAULT_SEQUENTIAL_TESTING_TUNING_PARAMETER
): Promise<ExperimentMetricAnalysis> {
  if (!rows || !rows.length) {
    return {
      unknownVariations: [],
      multipleExposures: 0,
      dimensions: [],
    };
  }

  const variationIdMap: { [key: string]: number } = {};
  variations.map((v, i) => {
    variationIdMap[v.id] = i;
  });

  logger.error("GRAB THE FOLLOWING AND SEND TO JAMES");
  logger.error(`
from gbstats.gbstats import (
  diff_for_daily_time_series,
  detect_unknown_variations,
  analyze_metric_df,
  get_metric_df,
  reduce_dimensionality,
  format_results
)
from gbstats.shared.constants import StatsEngine
import pandas as pd
import json

data = json.loads("""${JSON.stringify({
    var_id_map: variationIdMap,
    var_names: variations.map((v) => v.name),
    weights: variations.map((v) => v.weight),
    ignore_nulls: !!metric.ignoreNulls,
    inverse: !!metric.inverse,
    max_dimensions:
      dimension?.substring(0, 8) === "pre:date" ? 9999 : MAX_DIMENSIONS,
    rows,
  }).replace(/\\/g, "\\\\")}""", strict=False)

var_id_map = data['var_id_map']
var_names = data['var_names']
ignore_nulls = data['ignore_nulls']
inverse = data['inverse']
weights = data['weights']
max_dimensions = data['max_dimensions']

rows = pd.DataFrame(data['rows'])

unknown_var_ids = detect_unknown_variations(
  rows=rows,
  var_id_map=var_id_map
)

${
  dimension === "pre:datedaily" ? `rows = diff_for_daily_time_series(rows)` : ``
}

df = get_metric_df(
  rows=rows,
  var_id_map=var_id_map,
  var_names=var_names,
)

reduced = reduce_dimensionality(
  df=df, 
  max=max_dimensions
)

result = analyze_metric_df(
  df=reduced,
  weights=weights,
  inverse=inverse,
  engine=${
    statsEngine === "frequentist"
      ? "StatsEngine.FREQUENTIST"
      : "StatsEngine.BAYESIAN"
  },
  engine_config=${
    statsEngine === "frequentist" && sequentialTestingEnabled
      ? `{'sequential': True, 'sequential_tuning_parameter': ${sequentialTestingTuningParameter}}`
      : "{}"
  }
)

print(json.dumps({
  'unknownVariations': list(unknown_var_ids),
  'dimensions': format_results(result)
}, allow_nan=False))`);
  const result = await promisify(PythonShell.runString)(
    `
from gbstats.gbstats import (
  diff_for_daily_time_series,
  detect_unknown_variations,
  analyze_metric_df,
  get_metric_df,
  reduce_dimensionality,
  format_results
)
from gbstats.shared.constants import StatsEngine
import pandas as pd
import json

data = json.loads("""${JSON.stringify({
      var_id_map: variationIdMap,
      var_names: variations.map((v) => v.name),
      weights: variations.map((v) => v.weight),
      ignore_nulls: !!metric.ignoreNulls,
      inverse: !!metric.inverse,
      max_dimensions:
        dimension?.substring(0, 8) === "pre:date" ? 9999 : MAX_DIMENSIONS,
      rows,
    }).replace(/\\/g, "\\\\")}""", strict=False)

var_id_map = data['var_id_map']
var_names = data['var_names']
ignore_nulls = data['ignore_nulls']
inverse = data['inverse']
weights = data['weights']
max_dimensions = data['max_dimensions']

rows = pd.DataFrame(data['rows'])

unknown_var_ids = detect_unknown_variations(
  rows=rows,
  var_id_map=var_id_map
)

${
  dimension === "pre:datedaily" ? `rows = diff_for_daily_time_series(rows)` : ``
}

df = get_metric_df(
  rows=rows,
  var_id_map=var_id_map,
  var_names=var_names,
)

reduced = reduce_dimensionality(
  df=df, 
  max=max_dimensions
)

result = analyze_metric_df(
  df=reduced,
  weights=weights,
  inverse=inverse,
  engine=${
    statsEngine === "frequentist"
      ? "StatsEngine.FREQUENTIST"
      : "StatsEngine.BAYESIAN"
  },
  engine_config=${
    statsEngine === "frequentist" && sequentialTestingEnabled
      ? `{'sequential': True, 'sequential_tuning_parameter': ${sequentialTestingTuningParameter}}`
      : "{}"
  }
)

print(json.dumps({
  'unknownVariations': list(unknown_var_ids),
  'dimensions': format_results(result)
}, allow_nan=False))`,
    {}
  );

  let parsed: ExperimentMetricAnalysis;
  try {
    parsed = JSON.parse(result?.[0]);

    // Add multiple exposures
    parsed.multipleExposures =
      rows.filter((r) => r.variation === "__multiple__")?.[0]?.users || 0;
  } catch (e) {
    logger.error(e, "Failed to run stats model: " + result);
    throw e;
  }

  return parsed;
}

export async function analyzeExperimentResults({
  queryData,
  analysisSettings,
  snapshotSettings,
  variationNames,
  metricMap,
}: {
  queryData: QueryMap;
  analysisSettings: ExperimentSnapshotAnalysisSettings;
  snapshotSettings: ExperimentSnapshotSettings;
  variationNames?: string[];
  metricMap: Map<string, MetricInterface>;
}): Promise<ExperimentReportResults> {
  const metricRows: {
    metric: string;
    rows: ExperimentMetricQueryResponse;
  }[] = [];

  let unknownVariations: string[] = [];
  let multipleExposures = 0;

  // Everything done in a single query (Mixpanel, Google Analytics)
  // Need to convert to the same format as SQL rows
  if (queryData.has("results")) {
    const results = queryData.get("results");
    if (!results) throw new Error("Empty experiment results");
    const data = results.result as ExperimentResults;

    unknownVariations = data.unknownVariations;
    const byMetric: { [key: string]: ExperimentMetricQueryResponse } = {};
    data.dimensions.forEach((row) => {
      row.variations.forEach((v) => {
        Object.keys(v.metrics).forEach((metric) => {
          const stats = v.metrics[metric];
          byMetric[metric] = byMetric[metric] || [];
          byMetric[metric].push({
            dimension: row.dimension,
            variation:
              snapshotSettings.variations[v.variation]?.id || v.variation + "",
            users: stats.count,
            count: stats.count,
            statistic_type: "mean", // no ratio in mixpanel or GA
            main_metric_type: stats.metric_type,
            main_sum: stats.main_sum,
            main_sum_squares: stats.main_sum_squares,
          });
        });
      });
    });

    Object.keys(byMetric).forEach((metric) => {
      metricRows.push({
        metric,
        rows: byMetric[metric],
      });
    });
  }
  // One query for each metric, can just use the rows directly from the query
  else {
    queryData.forEach((query, key) => {
      const metric = metricMap.get(key);
      if (!metric) return;

      metricRows.push({
        metric: key,
        rows: query.result as ExperimentMetricQueryResponse,
      });
    });
  }

  const dimensionMap: Map<string, ExperimentReportResultDimension> = new Map();
  await promiseAllChunks(
    metricRows.map((data) => {
      const metric = metricMap.get(data.metric);
      return async () => {
        if (!metric) return;
        const result = await analyzeExperimentMetric(
          snapshotSettings.variations.map((v, i) => ({
            ...v,
            name: variationNames?.[i] || v.id,
          })),
          metric,
          data.rows,
          analysisSettings.dimensions[0],
          analysisSettings.statsEngine,
          analysisSettings.sequentialTesting,
          analysisSettings.sequentialTestingTuningParameter
        );
        unknownVariations = unknownVariations.concat(result.unknownVariations);
        multipleExposures = Math.max(
          multipleExposures,
          result.multipleExposures
        );

        result.dimensions.forEach((row) => {
          const dim = dimensionMap.get(row.dimension) || {
            name: row.dimension,
            srm: 1,
            variations: [],
          };

          row.variations.forEach((v, i) => {
            const data = dim.variations[i] || {
              users: v.users,
              metrics: {},
            };
            data.users = Math.max(data.users, v.users);
            data.metrics[metric.id] = {
              ...v,
              buckets: [],
            };
            dim.variations[i] = data;
          });

          dimensionMap.set(row.dimension, dim);
        });
      };
    }),
    3
  );

  const dimensions = Array.from(dimensionMap.values());
  if (!dimensions.length) {
    dimensions.push({
      name: "All",
      srm: 1,
      variations: [],
    });
  } else {
    dimensions.forEach((dimension) => {
      // Calculate SRM
      dimension.srm = checkSrm(
        dimension.variations.map((v) => v.users),
        snapshotSettings.variations.map((v) => v.weight)
      );
    });
  }

  return {
    multipleExposures,
    unknownVariations: Array.from(new Set(unknownVariations)),
    dimensions,
  };
}
You could replace your stats.ts file with that
then search the logs in the container for "JAMES" and copy the code below it.
oops, just realized that the console.logs might not print in a place you can find. It's better to use logger.error()
I editted the code above again so you could copy and paste that.
c

cuddly-fireman-12522

07/20/2023, 11:49 AM
@numerous-parrot-15513
m

melodic-ability-82567

07/20/2023, 11:54 AM
Hi @white-fireman-22476, We have Growthbook running in AWS ECS. I can make changes to the mentioned file, but to apply them, we need to restart the services. How can we achieve this? However, I’m concerned that restarting the service may cause health check failure and potentially kill the container. Do you have any ideas on how to handle this situation?
w

white-fireman-22476

07/20/2023, 12:16 PM
hmm good point. Are there other services running on that container you care about, or are you concerned it would just wipe out the modified code and then we wouldn't get the logs we need?
so I wasn't running it in aws, but I was able to locally docker exec into the container, change a file within dist directory, then outside the container do a docker restart and it picked up the changes made in the dist directory. The restart happened within ms or two so I don't think any health checks would fail
So basically these steps: • Log into the ec2 instance that is running. • sudo su - • docker ps • docker exec -it <CONAINTER ID OF GROWTHBOOK> bash • apt-get update • apt-get vim • vim packages/back-end/dist/services/stats.js • Edit the file, then :wq (to write and quit vim) • exit • docker restart <CONAINTER ID OF GROWTHBOOK>
The file itself needs to be the compiled version of the file
Copy code
"use strict";
var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
    function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
    return new (P || (P = Promise))(function (resolve, reject) {
        function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
        function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
        function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
        step((generator = generator.apply(thisArg, _arguments || [])).next());
    });
};
Object.defineProperty(exports, "__esModule", { value: true });
exports.analyzeExperimentResults = exports.analyzeExperimentMetric = exports.MAX_DIMENSIONS = void 0;
const util_1 = require("util");
const python_shell_1 = require("python-shell");
const constants_1 = require("shared/constants");
const promise_1 = require("../util/promise");
const stats_1 = require("../util/stats");
const logger_1 = require("../util/logger");
exports.MAX_DIMENSIONS = 20;
function analyzeExperimentMetric(variations, metric, rows, dimension = null, statsEngine = constants_1.DEFAULT_STATS_ENGINE, sequentialTestingEnabled = false, sequentialTestingTuningParameter = constants_1.DEFAULT_SEQUENTIAL_TESTING_TUNING_PARAMETER) {
    var _a, _b;
    return __awaiter(this, void 0, void 0, function* () {
        if (!rows || !rows.length) {
            return {
                unknownVariations: [],
                multipleExposures: 0,
                dimensions: [],
            };
        }
        const variationIdMap = {};
        variations.map((v, i) => {
            variationIdMap[v.id] = i;
        });
        logger_1.logger.error("GRAB THE FOLLOWING AND SEND TO JAMES");
        logger_1.logger.error(`
from gbstats.gbstats import (
  diff_for_daily_time_series,
  detect_unknown_variations,
  analyze_metric_df,
  get_metric_df,
  reduce_dimensionality,
  format_results
)
from gbstats.shared.constants import StatsEngine
import pandas as pd
import json

data = json.loads("""${JSON.stringify({
            var_id_map: variationIdMap,
            var_names: variations.map((v) => v.name),
            weights: variations.map((v) => v.weight),
            ignore_nulls: !!metric.ignoreNulls,
            inverse: !!metric.inverse,
            max_dimensions: (dimension === null || dimension === void 0 ? void 0 : dimension.substring(0, 8)) === "pre:date" ? 9999 : exports.MAX_DIMENSIONS,
            rows,
        }).replace(/\\/g, "\\\\")}""", strict=False)

var_id_map = data['var_id_map']
var_names = data['var_names']
ignore_nulls = data['ignore_nulls']
inverse = data['inverse']
weights = data['weights']
max_dimensions = data['max_dimensions']

rows = pd.DataFrame(data['rows'])

unknown_var_ids = detect_unknown_variations(
  rows=rows,
  var_id_map=var_id_map
)

${dimension === "pre:datedaily" ? `rows = diff_for_daily_time_series(rows)` : ``}

df = get_metric_df(
  rows=rows,
  var_id_map=var_id_map,
  var_names=var_names,
)

reduced = reduce_dimensionality(
  df=df, 
  max=max_dimensions
)

result = analyze_metric_df(
  df=reduced,
  weights=weights,
  inverse=inverse,
  engine=${statsEngine === "frequentist"
            ? "StatsEngine.FREQUENTIST"
            : "StatsEngine.BAYESIAN"},
  engine_config=${statsEngine === "frequentist" && sequentialTestingEnabled
            ? `{'sequential': True, 'sequential_tuning_parameter': ${sequentialTestingTuningParameter}}`
            : "{}"}
)

print(json.dumps({
  'unknownVariations': list(unknown_var_ids),
  'dimensions': format_results(result)
}, allow_nan=False))`);
        const result = yield (0, util_1.promisify)(python_shell_1.PythonShell.runString)(`
from gbstats.gbstats import (
  diff_for_daily_time_series,
  detect_unknown_variations,
  analyze_metric_df,
  get_metric_df,
  reduce_dimensionality,
  format_results
)
from gbstats.shared.constants import StatsEngine
import pandas as pd
import json

data = json.loads("""${JSON.stringify({
            var_id_map: variationIdMap,
            var_names: variations.map((v) => v.name),
            weights: variations.map((v) => v.weight),
            ignore_nulls: !!metric.ignoreNulls,
            inverse: !!metric.inverse,
            max_dimensions: (dimension === null || dimension === void 0 ? void 0 : dimension.substring(0, 8)) === "pre:date" ? 9999 : exports.MAX_DIMENSIONS,
            rows,
        }).replace(/\\/g, "\\\\")}""", strict=False)
var_id_map = data['var_id_map']
var_names = data['var_names']
ignore_nulls = data['ignore_nulls']
inverse = data['inverse']
weights = data['weights']
max_dimensions = data['max_dimensions']

rows = pd.DataFrame(data['rows'])

unknown_var_ids = detect_unknown_variations(
  rows=rows,
  var_id_map=var_id_map
)

${dimension === "pre:datedaily" ? `rows = diff_for_daily_time_series(rows)` : ``}

df = get_metric_df(
  rows=rows,
  var_id_map=var_id_map,
  var_names=var_names,
)

reduced = reduce_dimensionality(
  df=df, 
  max=max_dimensions
)

result = analyze_metric_df(
  df=reduced,
  weights=weights,
  inverse=inverse,
  engine=${statsEngine === "frequentist"
            ? "StatsEngine.FREQUENTIST"
            : "StatsEngine.BAYESIAN"},
  engine_config=${statsEngine === "frequentist" && sequentialTestingEnabled
            ? `{'sequential': True, 'sequential_tuning_parameter': ${sequentialTestingTuningParameter}}`
            : "{}"}
)

print(json.dumps({
  'unknownVariations': list(unknown_var_ids),
  'dimensions': format_results(result)
}, allow_nan=False))`, {});
        let parsed;
        try {
            parsed = JSON.parse(result === null || result === void 0 ? void 0 : result[0]);
            // Add multiple exposures
            parsed.multipleExposures =
                ((_b = (_a = rows.filter((r) => r.variation === "__multiple__")) === null || _a === void 0 ? void 0 : _a[0]) === null || _b === void 0 ? void 0 : _b.users) || 0;
        }
        catch (e) {
            logger_1.logger.error(e, "Failed to run stats model: " + result);
            throw e;
        }
        return parsed;
    });
}
exports.analyzeExperimentMetric = analyzeExperimentMetric;
function analyzeExperimentResults({ queryData, analysisSettings, snapshotSettings, variationNames, metricMap, }) {
    return __awaiter(this, void 0, void 0, function* () {
        const metricRows = [];
        let unknownVariations = [];
        let multipleExposures = 0;
        // Everything done in a single query (Mixpanel, Google Analytics)
        // Need to convert to the same format as SQL rows
        if (queryData.has("results")) {
            const results = queryData.get("results");
            if (!results)
                throw new Error("Empty experiment results");
            const data = results.result;
            unknownVariations = data.unknownVariations;
            const byMetric = {};
            data.dimensions.forEach((row) => {
                row.variations.forEach((v) => {
                    Object.keys(v.metrics).forEach((metric) => {
                        var _a;
                        const stats = v.metrics[metric];
                        byMetric[metric] = byMetric[metric] || [];
                        byMetric[metric].push({
                            dimension: row.dimension,
                            variation: ((_a = snapshotSettings.variations[v.variation]) === null || _a === void 0 ? void 0 : _a.id) || v.variation + "",
                            users: stats.count,
                            count: stats.count,
                            statistic_type: "mean",
                            main_metric_type: stats.metric_type,
                            main_sum: stats.main_sum,
                            main_sum_squares: stats.main_sum_squares,
                        });
                    });
                });
            });
            Object.keys(byMetric).forEach((metric) => {
                metricRows.push({
                    metric,
                    rows: byMetric[metric],
                });
            });
        }
        // One query for each metric, can just use the rows directly from the query
        else {
            queryData.forEach((query, key) => {
                const metric = metricMap.get(key);
                if (!metric)
                    return;
                metricRows.push({
                    metric: key,
                    rows: query.result,
                });
            });
        }
        const dimensionMap = new Map();
        yield (0, promise_1.promiseAllChunks)(metricRows.map((data) => {
            const metric = metricMap.get(data.metric);
            return () => __awaiter(this, void 0, void 0, function* () {
                if (!metric)
                    return;
                const result = yield analyzeExperimentMetric(snapshotSettings.variations.map((v, i) => (Object.assign(Object.assign({}, v), { name: (variationNames === null || variationNames === void 0 ? void 0 : variationNames[i]) || v.id }))), metric, data.rows, analysisSettings.dimensions[0], analysisSettings.statsEngine, analysisSettings.sequentialTesting, analysisSettings.sequentialTestingTuningParameter);
                unknownVariations = unknownVariations.concat(result.unknownVariations);
                multipleExposures = Math.max(multipleExposures, result.multipleExposures);
                result.dimensions.forEach((row) => {
                    const dim = dimensionMap.get(row.dimension) || {
                        name: row.dimension,
                        srm: 1,
                        variations: [],
                    };
                    row.variations.forEach((v, i) => {
                        const data = dim.variations[i] || {
                            users: v.users,
                            metrics: {},
                        };
                        data.users = Math.max(data.users, v.users);
                        data.metrics[metric.id] = Object.assign(Object.assign({}, v), { buckets: [] });
                        dim.variations[i] = data;
                    });
                    dimensionMap.set(row.dimension, dim);
                });
            });
        }), 3);
        const dimensions = Array.from(dimensionMap.values());
        if (!dimensions.length) {
            dimensions.push({
                name: "All",
                srm: 1,
                variations: [],
            });
        }
        else {
            dimensions.forEach((dimension) => {
                // Calculate SRM
                dimension.srm = (0, stats_1.checkSrm)(dimension.variations.map((v) => v.users), snapshotSettings.variations.map((v) => v.weight));
            });
        }
        return {
            multipleExposures,
            unknownVariations: Array.from(new Set(unknownVariations)),
            dimensions,
        };
    });
}
exports.analyzeExperimentResults = analyzeExperimentResults;
//# sourceMappingURL=stats.js.map
17 Views