Caution

You are not reading the most current version of the documentation. If you want up-to-date information, please have a look at 4.4 .

Tracing

Motivation

Tracing is a ScyllaDB tool that is meant to help debugging and analysing the internal flows in the server. One example of such a flow is a CQL request processing.

How to enable?

cqlsh

Users of a cqlsh may enable Tracing with TRACING ON|OFF command:

cqlsh> TRACING ON
Now Tracing is enabled
cqlsh> INSERT into keyspace1.standard1  (key, "C0") VALUES (0x12345679, bigintAsBlob(123456));

Tracing session: 227aff60-4f21-11e6-8835-000000000000

 activity                                                                                         | timestamp                  | source    | source_elapsed
--------------------------------------------------------------------------------------------------+----------------------------+-----------+----------------
                                                                               Execute CQL3 query | 2016-07-21 11:57:21.238000 | 127.0.0.2 |              0
                                                                    Parsing a statement [shard 1] | 2016-07-21 11:57:21.238335 | 127.0.0.2 |              1
                                                                 Processing a statement [shard 1] | 2016-07-21 11:57:21.238405 | 127.0.0.2 |             71
 Creating write handler for token: 2309717968349690594 natural: {127.0.0.1} pending: {} [shard 1] | 2016-07-21 11:57:21.238433 | 127.0.0.2 |             99
                                 Creating write handler with live: {127.0.0.1} dead: {} [shard 1] | 2016-07-21 11:57:21.238439 | 127.0.0.2 |            105
                                                       Sending a mutation to /127.0.0.1 [shard 1] | 2016-07-21 11:57:21.238490 | 127.0.0.2 |            156
                                                       Message received from /127.0.0.2 [shard 0] | 2016-07-21 11:57:21.238562 | 127.0.0.1 |             17
                                                    Sending mutation_done to /127.0.0.2 [shard 0] | 2016-07-21 11:57:21.238658 | 127.0.0.1 |            113
                                                              Mutation handling is done [shard 0] | 2016-07-21 11:57:21.238675 | 127.0.0.1 |            130
                                                         Got a response from /127.0.0.1 [shard 1] | 2016-07-21 11:57:21.238950 | 127.0.0.2 |            616
                                                        Mutation successfully completed [shard 1] | 2016-07-21 11:57:21.238958 | 127.0.0.2 |            624
                                                   Done processing - preparing a result [shard 1] | 2016-07-21 11:57:21.238962 | 127.0.0.2 |            628
                                                                                 Request complete | 2016-07-21 11:57:21.238639 | 127.0.0.2 |            639

Note how source_elapsed starts over on 127.0.0.1 when execution gets there.

The raw tracing data looks as follows:

cqlsh> select * from system_traces.sessions  where session_id=227aff60-4f21-11e6-8835-000000000000;

 session_id                           | client    | command | coordinator | duration | parameters                                                                                                                                                                                                                       | request            | started_at
--------------------------------------+-----------+---------+-------------+----------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+--------------------+---------------------------------
 227aff60-4f21-11e6-8835-000000000000 | 127.0.0.1 |   QUERY |   127.0.0.2 |      639 | {'consistency_level': 'ONE', 'page_size': '100', 'query': 'INSERT into keyspace1.standard1  (key, "C0") VALUES (0x12345679, bigintAsBlob(123456));', 'serial_consistency_level': 'SERIAL', 'user_timestamp': '1469091441238107'} | Execute CQL3 query | 2016-07-21 08:57:21.238000+0000

(1 rows)
cqlsh> select * from system_traces.events  where session_id=227aff60-4f21-11e6-8835-000000000000;

 session_id                           | event_id                             | activity                                                                               | source    | source_elapsed | thread
--------------------------------------+--------------------------------------+----------------------------------------------------------------------------------------+-----------+----------------+--------
 227aff60-4f21-11e6-8835-000000000000 | 227b0c74-4f21-11e6-8835-000000000000 |                                                                    Parsing a statement | 127.0.0.2 |              1 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b0f34-4f21-11e6-8835-000000000000 |                                                                 Processing a statement | 127.0.0.2 |             71 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b1047-4f21-11e6-8835-000000000000 | Creating write handler for token: 2309717968349690594 natural: {127.0.0.1} pending: {} | 127.0.0.2 |             99 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b1087-4f21-11e6-8835-000000000000 |                                 Creating write handler with live: {127.0.0.1} dead: {} | 127.0.0.2 |            105 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b1284-4f21-11e6-8835-000000000000 |                                                       Sending a mutation to /127.0.0.1 | 127.0.0.2 |            156 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b1559-4f21-11e6-bf08-000000000000 |                                                       Message received from /127.0.0.2 | 127.0.0.1 |             17 | shard 0
 227aff60-4f21-11e6-8835-000000000000 | 227b1915-4f21-11e6-bf08-000000000000 |                                                    Sending mutation_done to /127.0.0.2 | 127.0.0.1 |            113 | shard 0
 227aff60-4f21-11e6-8835-000000000000 | 227b19bd-4f21-11e6-bf08-000000000000 |                                                              Mutation handling is done | 127.0.0.1 |            130 | shard 0
 227aff60-4f21-11e6-8835-000000000000 | 227b247e-4f21-11e6-8835-000000000000 |                                                         Got a response from /127.0.0.1 | 127.0.0.2 |            616 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b24ca-4f21-11e6-8835-000000000000 |                                                        Mutation successfully completed | 127.0.0.2 |            624 | shard 1
 227aff60-4f21-11e6-8835-000000000000 | 227b24f2-4f21-11e6-8835-000000000000 |                                                   Done processing - preparing a result | 127.0.0.2 |            628 | shard 1

(11 rows)
cqlsh>

Probabilistic tracing

Tracing implies a significant performance penalty on a cluster when it’s enabled. Therefore if we want to enable tracing for some on going workload we don’t want to enable it for every request but rather for some (small) portion of them. This may be achieved using the so called probabilistic tracing, which would randomly choose a request to be traced with some defined probability.

For instance, if we want to trace 0.01% or all queries in the cluster we shell set a probabilistic tracing with the probability 0.0001:

$ nodetool settraceprobability 0.0001

How traces are stored?

Traces are stored in a system_traces keyspace for 24 hours, which consists of 2 tables with replication factor of 2:

CREATE TABLE system_traces.events (
    session_id uuid,
    event_id timeuuid,
    activity text,
    source inet,
    source_elapsed int,
    thread text,
    PRIMARY KEY (session_id, event_id)
)

CREATE TABLE system_traces.sessions (
    session_id uuid PRIMARY KEY,
    client inet,
    command text,
    coordinator inet,
    duration int,
    parameters map<text, text>,
    request text,
    started_at timestamp
)

Traces are created in a context of a tracing session. For instance, if we trace an INSERT CQL command, a tracing session with a unique ID (session_id column in the tables above) will be created and all trace points hit during the execution will be stored in a context of this session. And this defines the format in which tracing data is stored:

  • sessions table contains a single row for each tracing session

  • events table contains a single row for each trace point.

If we need trace points for a specific session we may query events table for this session’s ID (see examples above).

Note that there is no guaranteed way to know when the tracing of a particular session has completed and events now contains the full trace of this request. In particular, the appearance of the session’s id in the sessions table does not indicate that the tracing of this session int events has completed. This session record is written by the coordinator of the request when it considers the “foreground” part of the request to be done (and the duration field measures the time the request spent in foreground mode), but replicas may continue to write to the events table about their work for this request. Moreover, even on a single-node setup the writes to sessions and events happen asynchronously, so the user cannot rely on their relative ordering.

events columns descripton

events columns are quite straight forward:

  • session_id: ID of a session this trace

  • event_id: ID of this specific trace entry

  • activity: a trace message

  • source: address of a Node where the trace entry has been created

  • source_elapsed: a number of microseconds passed since the beginning of the tracing session on a specific Node (see examples above)

  • thread: currently this contains a number of a shard on which this trace point has been taken

sessions columns description

  • session_id: ID of this tracing session

  • command: currently this may only have a “QUERY” value

  • client: address of a Client that has sent this query

  • coordinator: address of a coordinator that received this query from a Client

  • duration: the total duration of this tracing session

  • parameters: this map contains string pairs that describe the query which may include:

    • query string

    • consistency level

    • etc.

  • request: a short string describing the current query, like “Execute CQL3 query”

  • started_at: is a timestamp taken when tracing session has began

Slow queries logging

The motivation

Many times in real life installations one of the most important parameters of the system is the longest response time. Naturally, the shorter it is - the better. Therefore capturing the request that take a long time and understanding why it took it so long is a very critical and challenging task.

The solution

The “slow query logging” is a ScyllaDB feature that is going to greatly ease the debugging related to the long requests. When enabled it would record the queries with the handling time above the specified threshold. As a result there will be created a new record in a system_traces.node_slow_log table. All tracing records created in a context of this query on a Coordinator Node will be written as well. In addition, if handling on some replica takes too long, its traces are going to be stored too.

Thereby, when we detect a slow query we are going to get a valuable tracing data that would greatly help us to understand what was the query and why it took so long to complete.

How to enable and configure

By default slow query logging is disabled. There is a REST API that allows configuring and querying the current configuration of this feature. To set the parameters run: curl -X POST --header "Content-Type: application/json" --header "Accept: application/json" "http://<Node's address>:10000/storage_service/slow_query?enable=<true|false>&ttl=<in seconds>&threshold=<threshold in microseconds>"

e.g. in order to disable the feature on a Node with an address 127.0.0.1, set the ttl to 8600s and a threshold to 10000us:

curl -X POST --header "Content-Type: application/json" --header "Accept: application/json" "http://127.0.0.1:10000/storage_service/slow_query?enable=false&ttl=8600&threshold=10000"

To get the current configuration run: curl -X GET --header "Content-Type: application/json" --header "Accept: application/json" "http://<Node's address>:10000/storage_service/slow_query"

e.g. after the POST command above the query result will look as below:

$ curl -X GET --header "Content-Type: application/json" --header "Accept: application/json" "http://127.0.0.1:10000/storage_service/slow_query"
{"threshold": 10000, "enable": false, "ttl": 8600}
$

node_slow_log table

Schema

CREATE TABLE system_traces.node_slow_log (
    start_time timeuuid,
    node_ip inet,
    shard int,
    command text,
    date timestamp,
    duration int,
    parameters map<text, text>,
    session_id uuid,
    source_ip inet,
    table_names set<text>,
    username text,
    PRIMARY KEY (start_time, node_ip, shard)
)

Columns description

  • start_time and date: time when the query has began

  • node_ip: Address of a Coordinator Node

  • shard: shard ID on a Coordinator, where the query has been handled

  • command: the query command, e.g. select * from my_ks.my_cf

  • duration: the duration of a query handling in microseconds

  • parameters: query parameters like a parameters column in a system_traces.sessions table

  • session_id: the corresponding Tracing session ID

  • source_ip: Address of a Client that sent this query

  • table_names: a list of tables used for this query, where applicable

  • username: a user name used for authentication with this query

Lightweight (fast) slow queries logging mode

The motivation

Natural desire is to run database with slow query tracing mode always enabled. But the implementation can’t detect early if the request will be slow before it got processed so it has to record all the tracing events before making a decision. Recording all the tracing events with all of its parameters during the request execution implies sufficient overhead. This lightweight mode or fast slow queries tracing mode offers a solution to this problem allowing low-overhead slow queries tracing.

The solution

The “fast slow query logging” is a ScyllaDB feature mode that is going to ease the debugging related to the long requests even further. It minimizes the tracing session related overhead to its minimum allowing it to be always enabled.

In a nutshell, this mode tracks only CQL statement and related request parameters. It effectively omits all the tracing events during the processing.

When enabled, it will work in the same way slow query tracing does besides that it will omit recording all the tracing events. So that it will not populate data to the system_traces.events table but it will populate trace session records for slow queries to all the rest: system_traces.sessions, system_traces.node_slow_log, etc.

Other tracing modes work as usual with that mode enabled.

How to enable and configure

By default fast slow query logging is disabled.

There is a REST API that allows configuring and querying the current configuration of this feature.

To request current state of the tracing run:

$ curl http://<node address>:10000/storage_service/slow_query

{"enable": false, "ttl": 86400, "threshold": 500000, "fast": false}

To enable fast slow query tracing run:

$ curl --request POST --header "Content-Type: application/json" --header "Accept: application/json" "http://<node address>:10000/storage_service/slow_query?enable=true&fast=true"

Normal slow query tracing can be enabled with:

$ curl --request POST --header "Content-Type: application/json" --header "Accept: application/json" "http://<node address>:10000/storage_service/slow_query?enable=true&fast=false"

Implementation details

Implementation introduces tracing._ignore_trace_events flag inside of the global sharded tracing service to track current tracing state. REST API calls tracing.set_ignore_trace_events() to modify its value.

tracing.create_session verifies if the trace_state is not a full_tracing request and tracing._ignore_trace_events is enabled than we can commit into omitting all the unnecessary data during tracing session.

Tracing session tracks tracing state by the means of trace_state_props::ignore_events that is kept in trace_state._state_props. In its turn tracing::make_trace_info does not create trace_info objects to prevent tracing of sub-events if tracing session has trace_state_props::ignore_events mode enabled.

Performance

We have found out that the fast slow queries tracing implies about 10 times less overhead on the requests processing than the normal slow query tracing in the best case hot path (100% cache hit rate prepared statements reads of a single row on 100% util shard).

In real production workloads we expect the effects to be almost completely invisible.

How to get query traces?

Each query tracing session gets a unique ID - session_id, which serves as a partition key for system_traces.sessions and system_traces.events tables.

Once one has this key the rest is trivial: SELECT * FROM system_traces.events WHERE session_id=<value>.

If we invoke tracing from the cqlsh using a TRACING ON command the session_id is printed with traces themselves and one can easily grab it and get the same traces later if needed using the query above.

If tracing was enabled using probabilistic tracing or with slow query log features (both described above) then one should use time based index tables: system_traces.sessions_time_idx and system_traces.node_slow_log_time_idx.

These indexes allow getting session_ids of all tracing sessions that were recorded in a specific time period.

For instance, to get session_ids of all tracing sessions that started between 2016-09-07 16:56:30-0000 and 2016-09-07 16:58:00-0000 one may run the following query:

SELECT session_id from system_traces.sessions_time_idx where minutes in ('2016-09-07 16:56:00-0000','2016-09-07 16:57:00-0000','2016-09-07 16:58:00-0000') and started_at > '2016-09-07 16:56:30-0000'
  • system_traces.node_slow_log_time_idx contains entries that correspond to queries traced only in the context of slow query logging.

  • system_traces.sessions_time_idx contains entries for all traced queries.

Index tables’ schemas are as follows:

CREATE TABLE system_traces.sessions_time_idx (
    minute timestamp,
    started_at timestamp,
    session_id uuid,
    PRIMARY KEY (minute, started_at, session_id))

CREATE TABLE system_traces.node_slow_log_time_idx (
    minute timestamp,
    started_at timestamp,
    session_id uuid,
    start_time timeuuid,
    node_ip inet,
    shard int,
    PRIMARY KEY (minute, started_at, session_id))

As you may notice each system_traces.node_slow_log_time_idx record contains system_traces.sessions, system_traces.events and system_traces.node_slow_log keys allowing to get the corresponding entries from each of these tables.