QueueUtil::hasScheduledJobOnMinute() optimisation

[vladimir.d]

Hi,

We have faced with an issue when a slow query executed against `job` table:

# Query_time: 12.193867 Lock_time: 0.000081 Rows_sent: 0 Rows_examined: 144349

SELECT job.id AS `id` FROM `job` WHERE job.scheduled_job_id = 'xxxxxyyyyxxxxiiiiixxxx' AND job.execute_time LIKE '2021-05-21 02:14:%' AND job.deleted = 0 LIMIT 0, 1;

There are 1350000 records in the table and the query takes about 10-14 seconds.

After a quick research we found a drawback and found a way to optimise the code - see the screenshot attached.



The proposed patch for version 6.1.7:

$ git diff

diff --git a/application/Espo/Core/Utils/Cron/Job.php b/application/Espo/Core/Utils/Cron/Job.php

index ee5b550d2..4900e3af5 100644

--- a/application/Espo/Core/Utils/Cron/Job.php

+++ b/application/Espo/Core/Utils/Cron/Job.php

@@ -165,13 +165,17 @@ class Job

public function hasScheduledJobOnMinute(string $scheduledJobId, string $time) : bool

{[INDENT]$dateObj = new DateTime($time);[/INDENT]
 
-         $timeWithoutSeconds = $dateObj->format('Y-m-d H:i:');

+        $minuteBegin = $dateObj->format('Y-m-d H:i:00');

+        $minuteEnd = $dateObj->format('Y-m-d H:i:59');



          $job = $this->getEntityManager()->getRepository('Job')

                ->select(['id'])

                ->where([

                    'scheduledJobId' => $scheduledJobId,

-                   'executeTime*' => $timeWithoutSeconds . '%',

+                  'AND' => [

+                          'executeTime>=' => $minuteBegin,

+                          'executeTime<=' => $minuteEnd

+                 ]

             ])

             ->findOne();



diff --git a/application/Espo/Resources/metadata/entityDefs/Job.json b/application/Espo/Resources/metadata/entityDefs/Job.json

index 71d45b0bf..125b5ecbd 100644

--- a/application/Espo/Resources/metadata/entityDefs/Job.json

+++ b/application/Espo/Resources/metadata/entityDefs/Job.json

@@ -107,6 +107,9 @@

     "textFilterFields": ["id", "name", "methodName", "serviceName", "scheduledJob"]

},

"indexes": {

+     "executeTimeExclusive": {

+     "columns": ["executeTime"]

+ },

"executeTime": {

       "columns": ["status", "executeTime"]

},

A patch for `master` branch:

$ git diff

diff --git a/application/Espo/Core/Job/QueueUtil.php b/application/Espo/Core/Job/QueueUtil.php

index 49f97ca98..14b5c744a 100644

--- a/application/Espo/Core/Job/QueueUtil.php

+++ b/application/Espo/Core/Job/QueueUtil.php

@@ -154,14 +154,18 @@ class QueueUtil

{

     $dateObj = new DateTime($time);



-    $timeWithoutSeconds = $dateObj->format('Y-m-d H:i:');

+   $minuteBegin = $dateObj->format('Y-m-d H:i:00');

+   $minuteEnd = $dateObj->format('Y-m-d H:i:59');



    $job = $this->entityManager

       ->getRDBRepository('Job')

       ->select(['id'])

       ->where([

           'scheduledJobId' => $scheduledJobId,

-          'executeTime*' => $timeWithoutSeconds . '%',

+         'AND' => [

+               'executeTime>=' => $minuteBegin,

+               'executeTime<=' => $minuteEnd

+         ]

     ])

     ->findOne();



diff --git a/application/Espo/Resources/metadata/entityDefs/Job.json b/application/Espo/Resources/metadata/entityDefs/Job.json

index 71d45b0bf..125b5ecbd 100644

--- a/application/Espo/Resources/metadata/entityDefs/Job.json

+++ b/application/Espo/Resources/metadata/entityDefs/Job.json

@@ -107,6 +107,9 @@

    "textFilterFields": ["id", "name", "methodName", "serviceName", "scheduledJob"]

},

"indexes": {

+     "executeTimeExclusive": {

+     "columns": ["executeTime"]

+ },

"executeTime": {

       "columns": ["status", "executeTime"]

},

Hope to see this patch in the upcoming release (6.1.8?).

Thank you!

[esforim]

Cool vladimir, not sure if you aware of their Github and you can probably do a Pull Request on it: https://github.com/espocrm/espocrm/pulls

By any chance are you using MariaDB?

[yuri]

Hi,

What database system do you use? I believe we didn't have performance issues here when we tested.

Is it necessary to add a new index? It's problematic to add indexes in hotfix version and we always try to do it only in minor/major releases.

[yuri]

What if we add an additional condition:

'status' => ['Pending', 'Ready', 'Running', 'Success'] 

Then an existing combined index should work out I believe.

[yuri]

I tested with an additional condition on status (along with executeTIme change) and it applied a proper index. So we can avoid adding an additional index. Could you test on your database whether the query performs fast?

[yuri]

The fix commit: https://github.com/espocrm/espocrm/c...f0e72db151139e


Please let me know if it executes fast on your environment. Thanks.

[vladimir.d]

yuri Thank you, the patch looks good but I cannot test it on our environment as we use EspoCRM Cloudron App and it's on read-only filesystem.

Could you `build` the query so I could test it?

[yuri]

SELECT id FROM job
WHERE
job.status IN ('Pending', 'Ready', 'Running', 'Success') AND
job.execute_time >= 'BEFORE' AND
job.execute_time <= 'AFTER' AND
job.deleted = 0

[vladimir.d]

Well done, it works quite fast, it takes about 0.1-0.5 seconds in our environment (cloudron app).
Hope the patch will appear in the nearest release.
Thank you