示例#1
0
文件: private.py 项目: ooni/api
def api_private_website_stats() -> Response:
    """TODO
    ---
    responses:
      '200':
        description: TODO
    """
    # uses_pg_index counters_day_cc_asn_input_idx a BRIN index was not used at
    # all, but BTREE on (measurement_start_day, probe_cc, probe_asn, input)
    # made queries go from full scan to 50ms
    url = request.args.get("input")
    probe_cc = validate_probe_cc_query_param()
    probe_asn = validate_probe_asn_query_param()

    s = """SELECT
        toDate(measurement_start_time) AS test_day,
        countIf(anomaly = 't') as anomaly_count,
        countIf(confirmed = 't') as confirmed_count,
        countIf(msm_failure = 't') as failure_count,
        COUNT() AS total_count
        FROM fastpath
        WHERE measurement_start_time >= today() - interval '31 day'
        AND measurement_start_time < today()
        AND probe_cc =  :probe_cc
        AND probe_asn = :probe_asn
        AND input = :input
        GROUP BY test_day ORDER BY test_day
    """
    d = {"probe_cc": probe_cc, "probe_asn": probe_asn, "input": url}
    results = query_click(sql.text(s), d)
    return cachedjson("0s", results=results)
示例#2
0
文件: private.py 项目: cyBerta/api
def api_private_test_names():
    """TODO
    ---
    responses:
      '200':
        description: TODO
    """
    TEST_NAMES = {
        "bridge_reachability": "Bridge Reachability",
        "dash": "DASH",
        "dns_consistency": "DNS Consistency",
        "facebook_messenger": "Facebook Messenger",
        "http_header_field_manipulation": "HTTP Header Field Manipulation",
        "http_host": "HTTP Host",
        "http_invalid_request_line": "HTTP Invalid Request Line",
        "http_requests": "HTTP Requests",
        "meek_fronted_requests_test": "Meek Fronted Requests",
        "multi_protocol_traceroute": "Multi Protocol Traceroute",
        "ndt": "NDT",
        "psiphon": "Psiphon",
        "tcp_connect": "TCP Connect",
        "telegram": "Telegram",
        "tor": "Tor",
        "vanilla_tor": "Vanilla Tor",
        "web_connectivity": "Web Connectivity",
        "whatsapp": "WhatsApp",
        "riseupvpn": "RiseUp VPN",
    }
    test_names = [{"id": k, "name": v} for k, v in TEST_NAMES.items()]
    return cachedjson(1, test_names=test_names)
示例#3
0
文件: private.py 项目: ooni/api
def check_report_id() -> Response:
    """Legacy. Used to check if a report_id existed in the fastpath table.
    Used by https://github.com/ooni/probe/issues/1034
    ---
    produces:
      - application/json
    parameters:
      - name: report_id
        in: query
        type: string
    responses:
      200:
        description: Always returns True.
        schema:
          type: object
          properties:
            v:
              type: integer
              description: version number of this response
            found:
              type: boolean
              description: True
          example: { "found": true, "v": 0 }

    """
    return cachedjson("1s", v=0, found=True)
示例#4
0
文件: private.py 项目: ooni/api
def api_private_global_by_month() -> Response:
    """Provide global summary of measurements
    Sources: global_by_month db table
    ---
    responses:
      '200':
        description: JSON struct TODO
    """
    q = """SELECT
        COUNT(DISTINCT probe_asn) AS networks_by_month,
        COUNT(DISTINCT probe_cc) AS countries_by_month,
        COUNT() AS measurements_by_month,
        toStartOfMonth(measurement_start_time) AS month
        FROM fastpath
        WHERE measurement_start_time > toStartOfMonth(today() - interval 2 year)
        AND measurement_start_time < toStartOfMonth(today() + interval 1 month)
        GROUP BY month ORDER BY month
    """
    rows = query_click(sql.text(q), {})
    rows = list(rows)

    n = [{"date": r["month"], "value": r["networks_by_month"]} for r in rows]
    c = [{"date": r["month"], "value": r["countries_by_month"]} for r in rows]
    m = [{"date": r["month"], "value": r["measurements_by_month"]} for r in rows]
    expand_dates(n)
    expand_dates(c)
    expand_dates(m)
    return cachedjson(
        "1d", networks_by_month=n, countries_by_month=c, measurements_by_month=m
    )
示例#5
0
文件: private.py 项目: ooni/api
def api_private_website_network_tests() -> Response:
    """TODO
    ---
    parameters:
      - name: probe_cc
        in: query
        type: string
        description: The two letter country code
    responses:
      '200':
        description: TODO
    """
    probe_cc = validate_probe_cc_query_param()
    s = """SELECT
        COUNT() AS count,
        probe_asn
        FROM fastpath
        WHERE
            measurement_start_time >= today() - interval '31 day'
        AND measurement_start_time < today()
        AND probe_cc = :probe_cc
        GROUP BY probe_asn
        ORDER BY count DESC
        """
    results = query_click(sql.text(s), {"probe_cc": probe_cc})
    return cachedjson("0s", results=results)
示例#6
0
文件: private.py 项目: ooni/api
def api_private_country_overview() -> Response:
    """Country-specific overview
    ---
    responses:
      '200':
        description: {
        "first_bucket_date":"2012-12-01",
        "measurement_count":6659891,
        "network_count":333}
    """
    # TODO: add circumvention_tools_blocked im_apps_blocked
    # middlebox_detected_networks websites_confirmed_blocked
    probe_cc = validate_probe_cc_query_param()
    s = """SELECT
        toDate(MIN(measurement_start_time)) AS first_bucket_date,
        COUNT() AS measurement_count,
        COUNT(DISTINCT probe_asn) AS network_count
        FROM fastpath
        WHERE probe_cc = :probe_cc
        AND measurement_start_time > '2012-12-01'
    """
    r = query_click_one_row(sql.text(s), {"probe_cc": probe_cc})
    assert r
    # FIXME: websites_confirmed_blocked
    return cachedjson("1d", **r)
示例#7
0
文件: private.py 项目: cyBerta/api
def api_private_country_overview():
    """Country-specific overview
    ---
    responses:
      '200':
        description: {
        "first_bucket_date":"2012-12-01",
        "measurement_count":6659891,
        "network_count":333}
    """
    # TODO: add circumvention_tools_blocked im_apps_blocked
    # middlebox_detected_networks websites_confirmed_blocked
    probe_cc = validate_probe_cc_query_param()
    # OLAP-use-case
    s = sql.text("""SELECT
        MIN(measurement_start_day) AS first_bucket_date,
        SUM(measurement_count) AS measurement_count,
        COUNT(DISTINCT probe_asn) AS network_count
        FROM counters_asn_noinput
        WHERE probe_cc = :probe_cc
    """)
    q = current_app.db_session.execute(s, {"probe_cc": probe_cc})
    r = q.fetchone()
    # NOTE: we are hardcoding a fixed first_bucket_date
    # FIXME: websites_confirmed_blocked
    return cachedjson(
        24,
        first_bucket_date=r.first_bucket_date,
        measurement_count=r.measurement_count,
        network_count=r.network_count,
    )
示例#8
0
文件: private.py 项目: ooni/api
def api_private_network_stats() -> Response:
    """Network speed statistics - not implemented
    ---
    parameters:
      - name: probe_cc
        in: query
        type: string
        description: The two letter country code
    responses:
      '200':
        description: TODO
    """
    # TODO: implement the stats from NDT in fastpath and then here
    probe_cc = validate_probe_cc_query_param()

    return cachedjson(
        "1d",
        {
            "metadata": {
                "current_page": 1,
                "limit": 10,
                "next_url": None,
                "offset": 0,
                "total_count": 0,
            },
            "results": [],
        },
    )
示例#9
0
def list_collectors() -> Response:
    """Probe Services: list collectors
    ---
    responses:
      '200':
        description: List available collectors
    """
    # TODO load from configuration file
    j = [
        {"address": "httpo://guegdifjy7bjpequ.onion", "type": "onion"},
        {"address": "https://ams-pg.ooni.org:443", "type": "https"},
        {
            "address": "https://dkyhjv0wpi2dk.cloudfront.net",
            "front": "dkyhjv0wpi2dk.cloudfront.net",
            "type": "cloudfront",
        },
        {"address": "httpo://guegdifjy7bjpequ.onion", "type": "onion"},
        {"address": "https://ams-pg.ooni.org:443", "type": "https"},
        {
            "address": "https://dkyhjv0wpi2dk.cloudfront.net",
            "front": "dkyhjv0wpi2dk.cloudfront.net",
            "type": "cloudfront",
        },
    ]
    return cachedjson("1h", j)
示例#10
0
文件: private.py 项目: ooni/api
def api_private_circumvention_runtime_stats() -> Response:
    """Runtime statistics on protocols used for circumvention,
    grouped by date, country, test_name.
    ---
    responses:
      200:
        description: List of dicts with keys probe_cc and cnt
    """
    q = """SELECT
        toDate(measurement_start_time) AS date,
        test_name,
        probe_cc,
        quantile(.5)(JSONExtractFloat(scores, 'extra', 'test_runtime')) AS p50,
        quantile(.9)(JSONExtractFloat(scores, 'extra', 'test_runtime')) AS p90,
        count() as cnt
    FROM fastpath
    WHERE test_name IN ['torsf', 'tor', 'stunreachability', 'psiphon','riseupvpn']
    AND measurement_start_time > today() - interval 6 month
    AND measurement_start_time < today() - interval 1 day
    AND JSONHas(scores, 'extra', 'test_runtime')
    GROUP BY date, probe_cc, test_name
    """
    try:
        r = query_click(sql.text(q), {})
        result = pivot_circumvention_runtime_stats(r)
        return cachedjson("1d", v=0, results=result)

    except Exception as e:
        return jsonify({"v": 0, "error": str(e)})
示例#11
0
文件: private.py 项目: cyBerta/api
def check_report_id():
    """Check if a report_id exists either in the fastpath table
    Used by https://github.com/ooni/probe/issues/1034
    ---
    responses:
      '200':
        description: TODO
    """
    report_id = request.args.get("report_id")
    s = sql.text("SELECT 1 FROM fastpath WHERE report_id = :rid LIMIT 1")
    try:
        q = current_app.db_session.execute(s, dict(rid=report_id))
        found = q.fetchone() is not None
        return cachedjson(5 / 3600, v=0, found=found)  # cache 5min

    except Exception as e:
        return cachedjson(0, v=0, error=str(e))
示例#12
0
def close_report(report_id) -> Response:
    """Probe Services: Close report
    ---
    responses:
      '200':
        description: Close a report
    """
    return cachedjson("1h")
示例#13
0
文件: private.py 项目: ooni/api
def api_private_vanilla_tor_stats() -> Response:
    """Tor statistics over ASN for a given CC
    ---
    parameters:
      - name: probe_cc
        in: query
        type: string
        minLength: 2
        required: true
    responses:
      '200':
        description: TODO
    """
    probe_cc = validate_probe_cc_query_param()
    blocked = 0
    nets = []
    s = """SELECT
        countIf(msm_failure = 't') as failure_count,
        toDate(MAX(measurement_start_time)) AS last_tested,
        probe_asn,
        COUNT() as total_count,
        total_count - countIf(anomaly = 't') AS success_count
        FROM fastpath
        WHERE measurement_start_time > today() - INTERVAL 6 MONTH
        AND probe_cc =  :probe_cc
        GROUP BY probe_asn
    """
    q = query_click(sql.text(s), {"probe_cc": probe_cc})
    extras = {
        "test_runtime_avg": None,
        "test_runtime_max": None,
        "test_runtime_min": None,
    }
    for n in q:
        n.update(extras)
        nets.append(n)
        if n["total_count"] > 5:
            if float(n["success_count"]) / float(n["total_count"]) < 0.6:
                blocked += 1

    if not nets:
        return cachedjson("0s", networks=[], notok_networks=0, last_tested=None)

    lt = max(n["last_tested"] for n in nets)
    return cachedjson("0s", networks=nets, notok_networks=blocked, last_tested=lt)
示例#14
0
文件: private.py 项目: ooni/api
def api_private_circumvention_stats_by_country() -> Response:
    """Aggregated statistics on protocols used for circumvention,
    grouped by country.
    ---
    responses:
      200:
        description: List of dicts with keys probe_cc and cnt
    """
    q = """SELECT probe_cc, COUNT(*) as cnt
        FROM fastpath
        WHERE measurement_start_time > today() - interval 6 month
        AND measurement_start_time < today() - interval 1 day
        AND test_name IN ['torsf', 'tor', 'stunreachability', 'psiphon','riseupvpn']
        GROUP BY probe_cc ORDER BY probe_cc
    """
    try:
        result = query_click(sql.text(q), {})
        return cachedjson("1d", v=0, results=result)

    except Exception as e:
        return cachedjson("0d", v=0, error=str(e))
示例#15
0
文件: private.py 项目: cyBerta/api
def api_private_im_networks():
    """TODO
    ---
    responses:
      '200':
        description: TODO
    """
    log = current_app.logger
    probe_cc = validate_probe_cc_query_param()
    test_names = [
        sql.literal_column("test_name") == t for t in TEST_GROUPS["im"]
    ]
    s = (
        select([
            sql.text("SUM(measurement_count) as msm_count"),
            sql.text("MAX(measurement_start_day) AS last_tested"),
            sql.text("probe_asn"),
            sql.text("test_name"),
        ]).where(
            and_(
                sql.text(
                    "measurement_start_day >= current_date - interval '31 day'"
                ),
                # We exclude the last day to wait for the pipeline
                sql.text("measurement_start_day < current_date"),
                sql.text("probe_cc = :probe_cc"),
                or_(*test_names),
            )).group_by(sql.text("test_name, probe_asn")).select_from(
                sql.table("counters_asn_noinput")).order_by(
                    sql.text("test_name, msm_count DESC")))

    results = {}
    q = current_app.db_session.execute(s, {"probe_cc": probe_cc})
    for r in q:
        e = results.get(
            r.test_name,
            {
                "anomaly_networks": [],
                "ok_networks": [],
                "last_tested": r.last_tested
            },
        )
        e["ok_networks"].append({
            "asn": r.probe_asn,
            "name": "",
            "total_count": r.msm_count,
            "last_tested": r.last_tested,
        })
        if e["last_tested"] < r.last_tested:
            e["last_tested"] = r.last_tested
        results[r.test_name] = e

    return cachedjson(1, **results)  # TODO caching
示例#16
0
文件: private.py 项目: cyBerta/api
def api_private_im_stats():
    """TODO
    ---
    responses:
      '200':
        description: TODO
    """
    test_name = request.args.get("test_name")
    if not test_name or test_name not in TEST_GROUPS["im"]:
        raise BadRequest("invalid test_name")

    probe_cc = validate_probe_cc_query_param()
    probe_asn = validate_probe_asn_query_param()
    probe_asn = int(probe_asn.replace("AS", ""))

    s = sql.text("""SELECT
        SUM(measurement_count) as total_count,
        measurement_start_day
        FROM counters_asn_noinput
        WHERE probe_cc = :probe_cc
        AND test_name = :test_name
        AND probe_asn = :probe_asn
        AND measurement_start_day >= current_date - interval '31 day'
        AND measurement_start_day < current_date
        GROUP BY measurement_start_day
        ORDER BY measurement_start_day
    """)

    query_params = {
        "probe_cc": probe_cc,
        "probe_asn": probe_asn,
        "test_name": test_name,
    }

    q = current_app.db_session.execute(s, query_params)

    tmp = {r.measurement_start_day: r for r in q}
    results = []
    days = [date.today() + timedelta(days=(d - 31)) for d in range(32)]
    for d in days:
        if d in tmp:
            e = {
                "test_day": isomid(tmp[d].measurement_start_day),
                "total_count": tmp[d].total_count,
            }
        else:
            e = {"test_day": isomid(d), "total_count": 0}

        e["anomaly_count"] = None
        results.append(e)

    return cachedjson(1, results=results)  # TODO caching
示例#17
0
文件: private.py 项目: cyBerta/api
def api_private_countries_by_month():
    """Countries count by month
    ---
    responses:
      '200':
        description: TODO
    """
    # OLAP-use-case
    cols = [sql.text("countries_by_month"), sql.text("month")]
    q = select(cols).select_from(sql.table("global_by_month"))
    q = current_app.db_session.execute(q)
    li = [dict(date=r[1], value=r[0]) for r in q]
    return cachedjson(24, li)
示例#18
0
文件: private.py 项目: cyBerta/api
def api_private_asn_by_month():
    """Network count by month
    ---
    responses:
      '200':
        description: [{"date":"2018-08-31","value":4411}, ... ]
    """
    # OLAP-use-case
    cols = [sql.text("networks_by_month"), sql.text("month")]
    q = select(cols).select_from(sql.table("global_by_month"))
    q = current_app.db_session.execute(q)
    li = [dict(date=r[1], value=r[0]) for r in q]
    return cachedjson(24, li)
示例#19
0
def check_report_id():
    """Check if a report_id exists in the fastpath table
    Used by https://github.com/ooni/probe/issues/1034
    ---
    parameters:
      - name: report_id
        in: query
        example: 20210208T162755Z_ndt_DZ_36947_n1_8swgXi7xNuRUyO9a
        type: string
        minLength: 10
        required: true
    responses:
      200:
        description: Check if a report_id exists in the fastpath table.
          Cached for a short time.
        schema:
          type: object
          properties:
            v:
              type: string
              description: version number of this response
            found:
              type: bool
              description: True if found
            error:
              type: string
              description: error message
          example: { "found": true, "v": 0 }

    """
    report_id = request.args.get("report_id")
    s = sql.text("SELECT 1 FROM fastpath WHERE report_id = :rid LIMIT 1")
    try:
        q = current_app.db_session.execute(s, dict(rid=report_id))
        found = q.fetchone() is not None
        return cachedjson(2 / 60, v=0, found=found)  # cache for 2min

    except Exception as e:
        return cachedjson(0, v=0, error=str(e))
示例#20
0
文件: private.py 项目: ooni/api
def api_private_im_stats() -> Response:
    """Instant messaging statistics
    ---
    responses:
      '200':
        description: TODO
    """
    test_name = request.args.get("test_name")
    if not test_name or test_name not in TEST_GROUPS["im"]:
        raise BadRequest("invalid test_name")

    probe_cc = validate_probe_cc_query_param()
    probe_asn = validate_probe_asn_query_param()
    probe_asn = int(probe_asn.replace("AS", ""))

    s = """SELECT
        COUNT() as total_count,
        toDate(measurement_start_time) AS test_day
        FROM fastpath
        WHERE probe_cc = :probe_cc
        AND test_name = :test_name
        AND probe_asn = :probe_asn
        AND measurement_start_time >= today() - interval '31 day'
        AND measurement_start_time < today()
        GROUP BY test_day
        ORDER BY test_day
    """
    query_params = {
        "probe_cc": probe_cc,
        "probe_asn": probe_asn,
        "test_name": test_name,
    }
    q = query_click(sql.text(s), query_params)
    tmp = {r["test_day"]: r for r in q}
    results = []
    days = [date.today() + timedelta(days=(d - 31)) for d in range(32)]
    for d in days:
        if d in tmp:
            test_day = isomid(tmp[d]["test_day"])
            total_count = tmp[d]["total_count"]
        else:
            test_day = isomid(d)
            total_count = 0
        e = dict(anomaly_count=None, test_day=test_day, total_count=total_count)
        results.append(e)

    return cachedjson("1h", results=results)  # TODO caching
示例#21
0
文件: private.py 项目: ooni/api
def api_private_im_networks() -> Response:
    """Instant messaging networks statistics
    ---
    responses:
      '200':
        description: TODO
    """
    probe_cc = validate_probe_cc_query_param()
    s = """SELECT
    COUNT() AS total_count,
    '' AS name,
    toDate(MAX(measurement_start_time)) AS last_tested,
    probe_asn,
    test_name
    FROM fastpath
    WHERE measurement_start_time >= today() - interval 31 day
    AND measurement_start_time < today()
    AND probe_cc = :probe_cc
    AND test_name IN :test_names
    GROUP BY test_name, probe_asn
    ORDER BY test_name ASC, total_count DESC
    """
    results: Dict[str, Dict] = {}
    test_names = sorted(TEST_GROUPS["im"])
    q = query_click(sql.text(s), {"probe_cc": probe_cc, "test_names": test_names})
    for r in q:
        e = results.get(
            r["test_name"],
            {
                "anomaly_networks": [],
                "ok_networks": [],
                "last_tested": r["last_tested"],
            },
        )
        e["ok_networks"].append(
            {
                "asn": r["probe_asn"],
                "name": "",
                "total_count": r["total_count"],
                "last_tested": r["last_tested"],
            }
        )
        if e["last_tested"] < r["last_tested"]:
            e["last_tested"] = r["last_tested"]
        results[r["test_name"]] = e

    return cachedjson("1h", **results)  # TODO caching
示例#22
0
文件: private.py 项目: ooni/api
def api_private_global_overview() -> Response:
    """Provide global summary of measurements
    Sources: global_stats db table
    ---
    responses:
      '200':
        description: JSON struct TODO
    """
    q = """SELECT
        COUNT(DISTINCT(probe_asn)) AS network_count,
        COUNT(DISTINCT probe_cc) AS country_count,
        COUNT(*) AS measurement_count
    FROM fastpath
    """
    r = query_click_one_row(q, {})
    assert r
    return cachedjson("1d", **r)
示例#23
0
文件: private.py 项目: ooni/api
def api_private_asn_by_month() -> Response:
    """Network count by month
    ---
    responses:
      '200':
        description: [{"date":"2018-08-31","value":4411}, ... ]
    """
    q = """SELECT
        COUNT(DISTINCT(probe_asn)) AS value,
        toStartOfMonth(measurement_start_time) AS date
    FROM fastpath
    WHERE measurement_start_time < toStartOfMonth(addMonths(now(), 1))
    AND measurement_start_time > toStartOfMonth(subtractMonths(now(), 24))
    GROUP BY date ORDER BY date
    """
    li = list(query_click(q, {}))
    expand_dates(li)
    return cachedjson("1d", li)
示例#24
0
文件: private.py 项目: cyBerta/api
def api_private_test_coverage():
    """Return number of measurements per day across test categories
    ---
    responses:
      '200':
          description: '{"network_coverage: [...], "test_coverage": [...]}'
    """
    # TODO: merge the two queries into one?
    # TODO: remove test categories or move aggregation to the front-end?
    # OLAP-use-case
    probe_cc = validate_probe_cc_query_param()
    test_groups = request.args.get("test_groups")
    if test_groups is not None:
        test_groups = test_groups.split(",")

    tc = get_recent_test_coverage(probe_cc)
    nc = get_recent_network_coverage(probe_cc, test_groups)
    return cachedjson(24, network_coverage=nc, test_coverage=tc)
示例#25
0
文件: private.py 项目: cyBerta/api
def api_private_global_overview():
    """Provide global summary of measurements
    Sources: global_stats db table
    ---
    responses:
      '200':
        description: JSON struct TODO
    """
    # OLAP-use-case
    #  CREATE MATERIALIZED VIEW global_stats AS
    #  SELECT
    #      COUNT(DISTINCT probe_asn) AS network_count,
    #      COUNT(DISTINCT probe_cc) AS country_count,
    #      SUM(measurement_count) AS measurement_count
    #  FROM counters;
    q = select([sql.text("*")]).select_from(sql.table("global_stats"))
    r = current_app.db_session.execute(q).fetchone()
    return cachedjson(24, **dict(r))
示例#26
0
文件: private.py 项目: ooni/api
def api_private_countries_by_month() -> Response:
    """Countries count by month
    ---
    responses:
      '200':
        description: TODO
    """
    q = """SELECT
        COUNT(DISTINCT(probe_cc)) AS value,
        toStartOfMonth(measurement_start_time) AS date
    FROM fastpath
    WHERE measurement_start_time < toStartOfMonth(addMonths(now(), 1))
    AND measurement_start_time > toStartOfMonth(subtractMonths(now(), 24))
    GROUP BY date ORDER BY date
    """
    li = list(query_click(q, {}))
    expand_dates(li)
    return cachedjson("1d", li)
示例#27
0
文件: private.py 项目: cyBerta/api
def api_private_website_stats():
    """TODO
    ---
    responses:
      '200':
        description: TODO
    """
    # uses_pg_index counters_day_cc_asn_input_idx a BRIN index was not used at
    # all, but BTREE on (measurement_start_day, probe_cc, probe_asn, input)
    # made queries go from full scan to 50ms
    url = request.args.get("input")
    probe_cc = validate_probe_cc_query_param()
    probe_asn = validate_probe_asn_query_param()

    # disable bitmapscan otherwise PG uses the BRIN indexes instead of BTREE
    s = sql.text("""
        SET enable_bitmapscan = off;
        SELECT
             measurement_start_day,
             SUM(anomaly_count) as anomaly_count,
             SUM(confirmed_count) as confirmed_count,
             SUM(failure_count) as failure_count,
             SUM(measurement_count) as measurement_count
             FROM counters
             WHERE measurement_start_day >= current_date - interval '31 day'
             AND measurement_start_day < current_date
             AND probe_cc =  :probe_cc
             AND probe_asn = :probe_asn
             AND input = :input
             GROUP BY measurement_start_day
        """)
    q = current_app.db_session.execute(s, {
        "probe_cc": probe_cc,
        "probe_asn": probe_asn,
        "input": url
    })
    results = [{
        "test_day": r.measurement_start_day,
        "anomaly_count": int(r.anomaly_count or 0),
        "confirmed_count": int(r.confirmed_count or 0),
        "failure_count": int(r.failure_count or 0),
        "total_count": int(r.measurement_count or 0),
    } for r in q]
    return cachedjson(1, results=results)
示例#28
0
文件: private.py 项目: cyBerta/api
def api_private_countries():
    """Summary of countries
    ---
    responses:
      '200':
        description: TODO
    """
    cols = [sql.text("measurement_count"), sql.text("probe_cc")]
    q = select(cols).select_from(sql.table("country_stats"))
    c = []
    for r in current_app.db_session.execute(q):
        try:
            name = lookup_country(r.probe_cc)
            c.append(
                dict(alpha_2=r.probe_cc, name=name, count=r.measurement_count))
        except KeyError:
            pass

    return cachedjson(24, countries=c)
示例#29
0
文件: private.py 项目: cyBerta/api
def api_private_vanilla_tor_stats():
    """Tor statistics over ASN for a given CC
    ---
    responses:
      '200':
        description: TODO
    """
    probe_cc = validate_probe_cc_query_param()

    s = sql.text("""SELECT
        SUM(failure_count) as failure_count,
        MAX(measurement_start_day) AS last_tested,
        probe_asn,
        SUM(anomaly_count) as anomaly_count,
        SUM(measurement_count) as total_count
        FROM counters_asn_noinput
        WHERE measurement_start_day > (current_date - interval '6 months')
        AND probe_cc =  :probe_cc
        GROUP BY probe_asn
    """)
    q = current_app.db_session.execute(s, {"probe_cc": probe_cc})

    nets = []
    blocked = 0
    for n in q:
        total_count = n.total_count or 0
        success_count = total_count - (n.anomaly_count or 0)
        nets.append({
            "failure_count": n.failure_count or 0,
            "last_tested": n.last_tested,
            "probe_asn": n.probe_asn,
            "success_count": success_count,
            "test_runtime_avg": None,
            "test_runtime_max": None,
            "test_runtime_min": None,
            "total_count": total_count,
        })
        if total_count > 5 and float(success_count) / float(total_count) < 0.6:
            blocked += 1

    lt = max(n["last_tested"] for n in nets)
    return cachedjson(0, networks=nets, notok_networks=blocked, last_tested=lt)
示例#30
0
文件: private.py 项目: cyBerta/api
def api_private_website_network_tests():
    """TODO
    ---
    responses:
      '200':
        description: TODO
    """
    probe_cc = validate_probe_cc_query_param()
    s = sql.text("""SELECT
        SUM(measurement_count) AS count,
        probe_asn
        FROM counters_asn_noinput
        WHERE
            measurement_start_day >= current_date - interval '31 day'
        AND measurement_start_day < current_date
        AND probe_cc = :probe_cc
        GROUP BY probe_asn
        ORDER BY count DESC
        """)
    q = current_app.db_session.execute(s, {"probe_cc": probe_cc})
    results = [dict(r) for r in q]
    return cachedjson(1, results=results)