Page MenuHomeSoftware Heritage
Differential D5978

Add a successive_visits counter to origin visit stats
ClosedPublic
Actions

Authored by ardumont on Jul 7 2021, 5:26 PM.

Details

Reviewers
olasd
douardda
vlorentz
Group Reviewers
Reviewers
Maniphest Tasks
T2345: Improve handling of recurrent loading tasks in scheduler
Commits
rDSCH1bcf84d5e66d: Add a successive_visits counter to origin visit stats
Summary

This maintains the number of successive visits resulting in the same status. This will
help implementing disabling of too many successive failed or not_found visits for a
given origin [1]

Related to T2345
Depends on D5956

[1] D5980

Supersedes D4895 (which I plan to commandeer and close)

Test Plan

tox

Diff Detail

Event Timeline

ardumont created this revision.Jul 7 2021, 5:26 PM
ardumont mentioned this in D4895: Add a successive_visits counter to OriginVisitStats.Jul 7 2021, 5:28 PM
swh-public-ci added a comment.Jul 7 2021, 5:29 PM

Build is green

Patch application report for D5978 (id=21541)

Could not rebase; Attempt merge onto 1006f0aee4...

Updating 1006f0a..7d71637
Fast-forward
 sql/updates/29.sql                         |  31 +++
 swh/scheduler/backend.py                   |  41 ++-
 swh/scheduler/interface.py                 |  19 ++
 swh/scheduler/journal_client.py            | 168 ++++++++++-
 swh/scheduler/model.py                     |  10 +
 swh/scheduler/sql/30-schema.sql            |  23 +-
 swh/scheduler/tests/test_api_client.py     |   2 +
 swh/scheduler/tests/test_journal_client.py | 429 ++++++++++++++++++++---------
 swh/scheduler/tests/test_scheduler.py      | 173 +++++++++++-
 9 files changed, 744 insertions(+), 152 deletions(-)
 create mode 100644 sql/updates/29.sql
Changes applied before test
commit 7d71637db7c6521b4cacbafd4e63e2bbd6432bd3
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to OriginVisitStats
    
    which (tries to) maintain the number of successive visits resulting in the same status.
    
    For example, if the last 3 visits are successful evenful ones, this field value will be
    3. Then, an uneventful visit coming right after the latest eventful one will reset this
    counter to one (last event is "uneventful" and only one such event happened
    successively).
    
    This will help when we will implement the disabling of too many failed visit attempt on
    a given origin
    
    Related to T2345

commit b02db7ce6222feeb5db7a7aff83a11c3a3697bd3
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Thu Jul 1 12:18:49 2021 +0200

    Introduce new scheduling policy to grab origins without last update
    
    This is in charge of scheduling origins without last update. This also updates the
    global queue position so the journal client can initialize correctly the next position
    per origin and visit type.
    
    Related to T2345

commit 8c4ae9f14d6abdca41a4f01b438310501ecb6259
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 16:00:01 2021 +0200

    journal_client: Compute next position for origin visit
    
    For origin without any last_update information [1], the journal client is now also in
    charge of moving their next position in the queue for rescheduling. Depending on their
    status, the next position offset and next_visit_queue_position are updated after each
    visit completes:
    
    - if the visit has failed, increase the next visit target by the minimal visit
      interval (to take into account transient loading issues)
    - if the visit is successful, and records some changes, decrease the visit interval
      index by 2 (visit the origin *way* more often).
    - if the visit is successful, and records no changes, increase the visit interval index
      by 1 (visit the origin less often).
    
    We then set the next visit target to its current value + the new visit interval
    multiplied by a random fudge factor (picked in the -/+ 10% range).
    
    The fudge factor allows the visits to spread out, avoiding "bursts" of loaded origins
    e.g. when a number of origins from a single hoster are processed at once.
    
    Note that the computations happen for all origins for simplicity and code maintenance
    but it will only be used by a new soon-to-be scheduling policy.
    
    [1] Lister cannot provide it for some reason.

commit cb1edf1ab24d1c8db5821578a7fb2633fab50ff4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 18:07:59 2021 +0200

    Introduce storage for the recurrent visit scheduler queue position

commit ec6e69f6415a007611c46f25e7c48e909a793d53
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:42:26 2021 +0200

    Start handling of recurrent loading tasks in scheduler
    
    This deals first and foremost with the next_position_offset update done by the scheduler
    journal client.

commit c486b28ece7c0b127fea10bbb4d7f5d1ad5c50ba
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 14:41:07 2021 +0200

    journal_client: Explicit docstring

commit 98f99b9fd457820dc2d4b5dab7e89cb8261a34a4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:39:40 2021 +0200

    journal_client: Only check last_* fields for some permutation tests
    
    In a future commit, we will add new fields whose values will be permutation dependent.

See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/410/ for more details.

Harbormaster completed remote builds in B22516: Diff 21541.Jul 7 2021, 5:29 PM
ardumont requested review of this revision.Jul 7 2021, 5:29 PM
anlambert added a subscriber: anlambert.Jul 7 2021, 5:38 PM

I got the idea but I am wondering if it would not be simpler to store the list of visit statuses directly in the model ?
This would allow finer processing for the scheduling policies, but this will be more costly to store.

swh/scheduler/journal_client.py
211

recent

ardumont added inline comments.Jul 7 2021, 5:47 PM
swh/scheduler/journal_client.py
211

lol i rebased a diff from david but i kept the typo, nice or what!

ardumont updated this revision to Diff 21544.Jul 7 2021, 5:47 PM

Fix typo

ardumont marked an inline comment as done.Jul 7 2021, 5:47 PM
swh-public-ci added a comment.Jul 7 2021, 5:50 PM

Build is green

Patch application report for D5978 (id=21544)

Could not rebase; Attempt merge onto 1006f0aee4...

Updating 1006f0a..5c7bd83
Fast-forward
 sql/updates/29.sql                         |  31 +++
 swh/scheduler/backend.py                   |  41 ++-
 swh/scheduler/interface.py                 |  19 ++
 swh/scheduler/journal_client.py            | 168 ++++++++++-
 swh/scheduler/model.py                     |  10 +
 swh/scheduler/sql/30-schema.sql            |  23 +-
 swh/scheduler/tests/test_api_client.py     |   2 +
 swh/scheduler/tests/test_journal_client.py | 429 ++++++++++++++++++++---------
 swh/scheduler/tests/test_scheduler.py      | 173 +++++++++++-
 9 files changed, 744 insertions(+), 152 deletions(-)
 create mode 100644 sql/updates/29.sql
Changes applied before test
commit 5c7bd832476083020f2ac71c4d3aaf415a579b50
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to OriginVisitStats
    
    which (tries to) maintain the number of successive visits resulting in the same status.
    
    For example, if the last 3 visits are successful evenful ones, this field value will be
    3. Then, an uneventful visit coming right after the latest eventful one will reset this
    counter to one (last event is "uneventful" and only one such event happened
    successively).
    
    This will help when we will implement the disabling of too many failed visit attempt on
    a given origin
    
    Related to T2345

commit b02db7ce6222feeb5db7a7aff83a11c3a3697bd3
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Thu Jul 1 12:18:49 2021 +0200

    Introduce new scheduling policy to grab origins without last update
    
    This is in charge of scheduling origins without last update. This also updates the
    global queue position so the journal client can initialize correctly the next position
    per origin and visit type.
    
    Related to T2345

commit 8c4ae9f14d6abdca41a4f01b438310501ecb6259
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 16:00:01 2021 +0200

    journal_client: Compute next position for origin visit
    
    For origin without any last_update information [1], the journal client is now also in
    charge of moving their next position in the queue for rescheduling. Depending on their
    status, the next position offset and next_visit_queue_position are updated after each
    visit completes:
    
    - if the visit has failed, increase the next visit target by the minimal visit
      interval (to take into account transient loading issues)
    - if the visit is successful, and records some changes, decrease the visit interval
      index by 2 (visit the origin *way* more often).
    - if the visit is successful, and records no changes, increase the visit interval index
      by 1 (visit the origin less often).
    
    We then set the next visit target to its current value + the new visit interval
    multiplied by a random fudge factor (picked in the -/+ 10% range).
    
    The fudge factor allows the visits to spread out, avoiding "bursts" of loaded origins
    e.g. when a number of origins from a single hoster are processed at once.
    
    Note that the computations happen for all origins for simplicity and code maintenance
    but it will only be used by a new soon-to-be scheduling policy.
    
    [1] Lister cannot provide it for some reason.

commit cb1edf1ab24d1c8db5821578a7fb2633fab50ff4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 18:07:59 2021 +0200

    Introduce storage for the recurrent visit scheduler queue position

commit ec6e69f6415a007611c46f25e7c48e909a793d53
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:42:26 2021 +0200

    Start handling of recurrent loading tasks in scheduler
    
    This deals first and foremost with the next_position_offset update done by the scheduler
    journal client.

commit c486b28ece7c0b127fea10bbb4d7f5d1ad5c50ba
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 14:41:07 2021 +0200

    journal_client: Explicit docstring

commit 98f99b9fd457820dc2d4b5dab7e89cb8261a34a4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:39:40 2021 +0200

    journal_client: Only check last_* fields for some permutation tests
    
    In a future commit, we will add new fields whose values will be permutation dependent.

See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/411/ for more details.

Harbormaster completed remote builds in B22519: Diff 21544.Jul 7 2021, 5:50 PM
ardumont edited the summary of this revision. (Show Details)Jul 7 2021, 5:51 PM
ardumont updated this revision to Diff 21545.Jul 7 2021, 5:51 PM
ardumont edited the summary of this revision. (Show Details)

reword commit message

swh-public-ci added a comment.Jul 7 2021, 5:54 PM

Build is green

Patch application report for D5978 (id=21545)

Could not rebase; Attempt merge onto 1006f0aee4...

Updating 1006f0a..4723b0c
Fast-forward
 sql/updates/29.sql                         |  31 +++
 swh/scheduler/backend.py                   |  41 ++-
 swh/scheduler/interface.py                 |  19 ++
 swh/scheduler/journal_client.py            | 168 ++++++++++-
 swh/scheduler/model.py                     |  10 +
 swh/scheduler/sql/30-schema.sql            |  23 +-
 swh/scheduler/tests/test_api_client.py     |   2 +
 swh/scheduler/tests/test_journal_client.py | 429 ++++++++++++++++++++---------
 swh/scheduler/tests/test_scheduler.py      | 173 +++++++++++-
 9 files changed, 744 insertions(+), 152 deletions(-)
 create mode 100644 sql/updates/29.sql
Changes applied before test
commit 4723b0c0e08e7c2dd942881c3f3dad78ec143e5f
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to OriginVisitStats
    
    This maintains the number of successive visits resulting in the same status.
    
    For example, when the last 3 visits are successful evenful ones, this field value will
    be 3. Then, an uneventful visit coming right after the latest eventful one will reset
    this counter to 1 (last event is "uneventful" and only one such event happened
    successively)...
    
    This will help when implementing disabling of too many failed visit attempts on a given
    origin.
    
    Related to T2345

commit b02db7ce6222feeb5db7a7aff83a11c3a3697bd3
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Thu Jul 1 12:18:49 2021 +0200

    Introduce new scheduling policy to grab origins without last update
    
    This is in charge of scheduling origins without last update. This also updates the
    global queue position so the journal client can initialize correctly the next position
    per origin and visit type.
    
    Related to T2345

commit 8c4ae9f14d6abdca41a4f01b438310501ecb6259
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 16:00:01 2021 +0200

    journal_client: Compute next position for origin visit
    
    For origin without any last_update information [1], the journal client is now also in
    charge of moving their next position in the queue for rescheduling. Depending on their
    status, the next position offset and next_visit_queue_position are updated after each
    visit completes:
    
    - if the visit has failed, increase the next visit target by the minimal visit
      interval (to take into account transient loading issues)
    - if the visit is successful, and records some changes, decrease the visit interval
      index by 2 (visit the origin *way* more often).
    - if the visit is successful, and records no changes, increase the visit interval index
      by 1 (visit the origin less often).
    
    We then set the next visit target to its current value + the new visit interval
    multiplied by a random fudge factor (picked in the -/+ 10% range).
    
    The fudge factor allows the visits to spread out, avoiding "bursts" of loaded origins
    e.g. when a number of origins from a single hoster are processed at once.
    
    Note that the computations happen for all origins for simplicity and code maintenance
    but it will only be used by a new soon-to-be scheduling policy.
    
    [1] Lister cannot provide it for some reason.

commit cb1edf1ab24d1c8db5821578a7fb2633fab50ff4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 18:07:59 2021 +0200

    Introduce storage for the recurrent visit scheduler queue position

commit ec6e69f6415a007611c46f25e7c48e909a793d53
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:42:26 2021 +0200

    Start handling of recurrent loading tasks in scheduler
    
    This deals first and foremost with the next_position_offset update done by the scheduler
    journal client.

commit c486b28ece7c0b127fea10bbb4d7f5d1ad5c50ba
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 14:41:07 2021 +0200

    journal_client: Explicit docstring

commit 98f99b9fd457820dc2d4b5dab7e89cb8261a34a4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:39:40 2021 +0200

    journal_client: Only check last_* fields for some permutation tests
    
    In a future commit, we will add new fields whose values will be permutation dependent.

See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/412/ for more details.

Harbormaster completed remote builds in B22520: Diff 21545.Jul 7 2021, 5:55 PM
ardumont added a comment.EditedJul 7 2021, 6:01 PM

I got the idea but I am wondering if it would not be simpler to store the list of visit statuses directly in the model ?

It's in the model ;) (but not in ListedOrigin if that's what you meant ;)

This would allow finer processing for the scheduling policies, but this will be more costly to store.

Hum, yes, i think that will be too much to store and i'm not sure what the other benefits would be
aside from the current need [1]. Also, that duplicates the information we already have in the archive?

[1] Right now, the need is a simple one, a counter to ease determine if too many failed
attempts occurred or not. If so, disable the origin in the model here (it should come soon, before the end of the week, i'd say ;)

anlambert added a comment.Jul 7 2021, 6:09 PM
In D5978#153850, @ardumont wrote:

I got the idea but I am wondering if it would not be simpler to store the list of visit statuses directly in the model ?

It's in the model ;) (but not in ListedOrigin if that's what you meant ;)

This would allow finer processing for the scheduling policies, but this will be more costly to store.

Hum, yes, i think that will be too much to store and i'm not sure what the other benefits would be
aside from the current need [1]. Also, that duplicates the information we already have in the archive?

[1] Right now, the need is a simple one, a counter to ease determine if too many failed
attempts occurred or not. If so, disable the origin in the model here (it will come during the week

Ack, I guess you will find the visit status associated to the counter based on the dates saved in the stats model ?

ardumont added a comment.Jul 7 2021, 7:26 PM
In D5978#153858, @anlambert wrote:
In D5978#153850, @ardumont wrote:

I got the idea but I am wondering if it would not be simpler to store the list of visit statuses directly in the model ?

It's in the model ;) (but not in ListedOrigin if that's what you meant ;)

This would allow finer processing for the scheduling policies, but this will be more costly to store.

Hum, yes, i think that will be too much to store and i'm not sure what the other benefits would be
aside from the current need [1]. Also, that duplicates the information we already have in the archive?

[1] Right now, the need is a simple one, a counter to ease determine if too many failed
attempts occurred or not. If so, disable the origin in the model here (it will come during the week

Ack, I guess you will find the visit status associated to the counter based on the dates saved in the stats model ?

yes, something like that ;)

ardumont retitled this revision from Add a successive_visits counter to OriginVisitStats to Add a successive_visits counter to origin visit stats.Jul 8 2021, 10:29 AM
ardumont edited the summary of this revision. (Show Details)
ardumont updated this revision to Diff 21551.Jul 8 2021, 10:29 AM

Refactor (to extract a function) and add tests around that function

swh-public-ci added a comment.Jul 8 2021, 10:32 AM

Build has FAILED

Patch application report for D5978 (id=21551)

Could not rebase; Attempt merge onto 1006f0aee4...

Updating 1006f0a..196254c
Fast-forward
 sql/updates/29.sql                         |  31 ++
 swh/scheduler/backend.py                   |  41 ++-
 swh/scheduler/interface.py                 |  19 ++
 swh/scheduler/journal_client.py            | 183 ++++++++++-
 swh/scheduler/model.py                     |  10 +
 swh/scheduler/sql/30-schema.sql            |  23 +-
 swh/scheduler/tests/test_api_client.py     |   2 +
 swh/scheduler/tests/test_journal_client.py | 473 +++++++++++++++++++++--------
 swh/scheduler/tests/test_scheduler.py      | 173 ++++++++++-
 9 files changed, 802 insertions(+), 153 deletions(-)
 create mode 100644 sql/updates/29.sql
Changes applied before test
commit 196254ce6b8e974d73ee7d3285dcd5f1ed32c017
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to origin visit stats
    
    This maintains the number of successive visits resulting in the same status. This will
    help when implementing the disabling of too many failed visit attempts for a given
    origin.
    
    Related to T2345

commit b02db7ce6222feeb5db7a7aff83a11c3a3697bd3
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Thu Jul 1 12:18:49 2021 +0200

    Introduce new scheduling policy to grab origins without last update
    
    This is in charge of scheduling origins without last update. This also updates the
    global queue position so the journal client can initialize correctly the next position
    per origin and visit type.
    
    Related to T2345

commit 8c4ae9f14d6abdca41a4f01b438310501ecb6259
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 16:00:01 2021 +0200

    journal_client: Compute next position for origin visit
    
    For origin without any last_update information [1], the journal client is now also in
    charge of moving their next position in the queue for rescheduling. Depending on their
    status, the next position offset and next_visit_queue_position are updated after each
    visit completes:
    
    - if the visit has failed, increase the next visit target by the minimal visit
      interval (to take into account transient loading issues)
    - if the visit is successful, and records some changes, decrease the visit interval
      index by 2 (visit the origin *way* more often).
    - if the visit is successful, and records no changes, increase the visit interval index
      by 1 (visit the origin less often).
    
    We then set the next visit target to its current value + the new visit interval
    multiplied by a random fudge factor (picked in the -/+ 10% range).
    
    The fudge factor allows the visits to spread out, avoiding "bursts" of loaded origins
    e.g. when a number of origins from a single hoster are processed at once.
    
    Note that the computations happen for all origins for simplicity and code maintenance
    but it will only be used by a new soon-to-be scheduling policy.
    
    [1] Lister cannot provide it for some reason.

commit cb1edf1ab24d1c8db5821578a7fb2633fab50ff4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 18:07:59 2021 +0200

    Introduce storage for the recurrent visit scheduler queue position

commit ec6e69f6415a007611c46f25e7c48e909a793d53
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:42:26 2021 +0200

    Start handling of recurrent loading tasks in scheduler
    
    This deals first and foremost with the next_position_offset update done by the scheduler
    journal client.

commit c486b28ece7c0b127fea10bbb4d7f5d1ad5c50ba
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 14:41:07 2021 +0200

    journal_client: Explicit docstring

commit 98f99b9fd457820dc2d4b5dab7e89cb8261a34a4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:39:40 2021 +0200

    journal_client: Only check last_* fields for some permutation tests
    
    In a future commit, we will add new fields whose values will be permutation dependent.

Link to build: https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/413/
See console output for more information: https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/413/console

Harbormaster failed remote builds in B22525: Diff 21551!Jul 8 2021, 10:32 AM
ardumont updated this revision to Diff 21552.Jul 8 2021, 10:35 AM

Fix

swh-public-ci added a comment.Jul 8 2021, 10:37 AM

Build is green

Patch application report for D5978 (id=21552)

Could not rebase; Attempt merge onto 1006f0aee4...

Updating 1006f0a..64aa445
Fast-forward
 sql/updates/29.sql                         |  31 ++
 swh/scheduler/backend.py                   |  41 ++-
 swh/scheduler/interface.py                 |  19 ++
 swh/scheduler/journal_client.py            | 181 ++++++++++-
 swh/scheduler/model.py                     |  10 +
 swh/scheduler/sql/30-schema.sql            |  23 +-
 swh/scheduler/tests/test_api_client.py     |   2 +
 swh/scheduler/tests/test_journal_client.py | 473 +++++++++++++++++++++--------
 swh/scheduler/tests/test_scheduler.py      | 173 ++++++++++-
 9 files changed, 801 insertions(+), 152 deletions(-)
 create mode 100644 sql/updates/29.sql
Changes applied before test
commit 64aa4458ddd2bb5bd9a913c17950732d962129e6
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to origin visit stats
    
    This maintains the number of successive visits resulting in the same status. This will
    help when implementing the disabling of too many failed visit attempts for a given
    origin.
    
    Related to T2345

commit b02db7ce6222feeb5db7a7aff83a11c3a3697bd3
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Thu Jul 1 12:18:49 2021 +0200

    Introduce new scheduling policy to grab origins without last update
    
    This is in charge of scheduling origins without last update. This also updates the
    global queue position so the journal client can initialize correctly the next position
    per origin and visit type.
    
    Related to T2345

commit 8c4ae9f14d6abdca41a4f01b438310501ecb6259
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 16:00:01 2021 +0200

    journal_client: Compute next position for origin visit
    
    For origin without any last_update information [1], the journal client is now also in
    charge of moving their next position in the queue for rescheduling. Depending on their
    status, the next position offset and next_visit_queue_position are updated after each
    visit completes:
    
    - if the visit has failed, increase the next visit target by the minimal visit
      interval (to take into account transient loading issues)
    - if the visit is successful, and records some changes, decrease the visit interval
      index by 2 (visit the origin *way* more often).
    - if the visit is successful, and records no changes, increase the visit interval index
      by 1 (visit the origin less often).
    
    We then set the next visit target to its current value + the new visit interval
    multiplied by a random fudge factor (picked in the -/+ 10% range).
    
    The fudge factor allows the visits to spread out, avoiding "bursts" of loaded origins
    e.g. when a number of origins from a single hoster are processed at once.
    
    Note that the computations happen for all origins for simplicity and code maintenance
    but it will only be used by a new soon-to-be scheduling policy.
    
    [1] Lister cannot provide it for some reason.

commit cb1edf1ab24d1c8db5821578a7fb2633fab50ff4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 18:07:59 2021 +0200

    Introduce storage for the recurrent visit scheduler queue position

commit ec6e69f6415a007611c46f25e7c48e909a793d53
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:42:26 2021 +0200

    Start handling of recurrent loading tasks in scheduler
    
    This deals first and foremost with the next_position_offset update done by the scheduler
    journal client.

commit c486b28ece7c0b127fea10bbb4d7f5d1ad5c50ba
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Tue Jun 29 14:41:07 2021 +0200

    journal_client: Explicit docstring

commit 98f99b9fd457820dc2d4b5dab7e89cb8261a34a4
Author: Antoine R. Dumont (@ardumont) <ardumont@softwareheritage.org>
Date:   Wed Jun 23 16:39:40 2021 +0200

    journal_client: Only check last_* fields for some permutation tests
    
    In a future commit, we will add new fields whose values will be permutation dependent.

See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/414/ for more details.

Harbormaster completed remote builds in B22526: Diff 21552.Jul 8 2021, 10:37 AM
ardumont added a child revision: D5980: journal_client: Disable origins when too many visited attempts failed.Jul 8 2021, 11:26 AM
ardumont added a comment.Jul 8 2021, 11:44 AM

... it will come before the end of the week

Ack, I guess you will find the visit status associated to the counter based on the dates saved in the stats model ?

yes, something like that ;)

done in D5980 ;)

ardumont mentioned this in T2345: Improve handling of recurrent loading tasks in scheduler.Jul 8 2021, 12:30 PM
ardumont planned changes to this revision.Jul 22 2021, 6:57 PM
ardumont updated this revision to Diff 21881.Aug 2 2021, 5:37 PM

Rebase on latest work

ardumont edited the summary of this revision. (Show Details)Aug 2 2021, 5:38 PM
swh-public-ci added a comment.Aug 2 2021, 5:40 PM

Build is green

Patch application report for D5978 (id=21881)

Rebasing onto 4fa29fe128...

Current branch diff-target is up to date.
Changes applied before test
commit cdc2af4733752ddccea01bdbd70b9805fbdaf6f1
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to origin visit stats
    
    This maintains the number of successive visits resulting in the same status. This will
    help implementing disabling of too many successive failed or not_found visits for a
    given origin.
    
    Related to T2345

See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/420/ for more details.

Harbormaster completed remote builds in B22846: Diff 21881.Aug 2 2021, 5:40 PM
ardumont updated this revision to Diff 21885.Aug 3 2021, 11:11 AM
ardumont edited the summary of this revision. (Show Details)

Start successive_visits to 0 like in the model declaration

swh-public-ci added a comment.Aug 3 2021, 11:13 AM

Build is green

Patch application report for D5978 (id=21885)

Rebasing onto 4fa29fe128...

First, rewinding head to replay your work on top of it...
Fast-forwarded diff-target to base-revision-422-D5978.
Changes applied before test



See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/422/ for more details.
Harbormaster completed remote builds in B22850: Diff 21885.Aug 3 2021, 11:13 AM
ardumont updated this revision to Diff 21888.Aug 3 2021, 11:13 AM
Use correct commit range...
swh-public-ci added a comment.Aug 3 2021, 11:21 AM
Build is green



Patch application report for D5978 (id=21888)



Rebasing onto 4fa29fe128...



Current branch diff-target is up to date.





Changes applied before test



commit d616934db615e6f53ea89c629a6b660bd24176e4
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to origin visit stats
    
    This maintains the number of successive visits resulting in the same status. This will
    help implementing disabling of too many successive failed or not_found visits for a
    given origin.
    
    Related to T2345



See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/425/ for more details.
Harbormaster completed remote builds in B22853: Diff 21888.Aug 3 2021, 11:21 AM
ardumont updated this revision to Diff 21889.Aug 3 2021, 11:21 AM
Add missing assertions on successive_visits
swh-public-ci added a comment.Aug 3 2021, 11:24 AM
Build is green



Patch application report for D5978 (id=21889)



Rebasing onto 4fa29fe128...



Current branch diff-target is up to date.





Changes applied before test



commit 015d16158df9a87cdea29d76a55381d6798ee4e3
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to origin visit stats
    
    This maintains the number of successive visits resulting in the same status. This will
    help implementing disabling of too many successive failed or not_found visits for a
    given origin.
    
    Related to T2345



See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/426/ for more details.
Harbormaster completed remote builds in B22855: Diff 21889.Aug 3 2021, 11:24 AM
vlorentz requested changes to this revision.Aug 3 2021, 11:59 AM
vlorentz added a subscriber: vlorentz.
Informally, if there is a single visit with a given status, then the number of successive visits with that status is 1; but with the semantics defined here, it's 0. This seems confusing to me; could you make counters start at 1 (unless there are no visits at all, in which case 0 is correct)
swh/scheduler/journal_client.py


222–237
I find this more readable


259
if increment_successive_visits == False resets the counter, it's probably not the best variable name.



What about same_visit_status?
This revision now requires changes to proceed.Aug 3 2021, 11:59 AM
ardumont added a comment.Aug 3 2021, 12:27 PM
Informally, if there is a single visit with a given status, then the number of

successive visits with that status is 1; but with the semantics defined here, it's 0.

This seems confusing to me; could you make counters start at 1 (unless there are no

visits at all, in which case 0 is correct)



yes, I sense it is confusing. Although starting the counter at 1 then resets it at 0

after when we switch to a new status is what's confusing to me...



Shouldn't we make that counter always start at 1 and then be reset to 1 instead?



My trail of thought is that if we have an entry in origin_visit_stats it means there is

at least one visit so 1 sound fine to me (it's also in accord with the start of your

sentence).



thoughts?
ardumont edited the summary of this revision. (Show Details)Aug 3 2021, 12:31 PM
vlorentz added a comment.Aug 3 2021, 12:39 PM


In D5978#156375, @ardumont wrote:


Informally, if there is a single visit with a given status, then the number of

successive visits with that status is 1; but with the semantics defined here, it's 0.

This seems confusing to me; could you make counters start at 1 (unless there are no

visits at all, in which case 0 is correct)



yes, I sense it is confusing. Although starting the counter at 1 then resets it at 0

after when we switch to a new status is what's confusing to me...



Shouldn't we make that counter always start at 1 and then be reset to 1 instead?





That's what I said. It's 0 before we have any visit, but as soon as we have a visit, it will never go back to 0 again.



My trail of thought is that if we have an entry in origin_visit_stats it means there is

at least one visit so 1 sound fine to me (it's also in accord with the start of your

sentence).



Then 0 is never recorded in the DB, fine with me.
ardumont updated this revision to Diff 21894.Aug 3 2021, 12:51 PM
ardumont marked 2 inline comments as done.
Adapt according to review:



    

  • start increment to 1 then adapt tests
    • 

  • rename variable to a more meaningful name
    • 

swh-public-ci added a comment.Aug 3 2021, 12:53 PM
Build is green



Patch application report for D5978 (id=21894)



Rebasing onto 4fa29fe128...



Current branch diff-target is up to date.





Changes applied before test



commit 1bcf84d5e66d02c006698a89d2571911d3fd0764
Author: David Douard <david.douard@sdfa3.org>
Date:   Wed Jul 7 16:55:57 2021 +0200

    Add a successive_visits counter to origin visit stats
    
    This maintains the number of successive visits resulting in the same status. This will
    help implementing disabling of too many successive failed or not_found visits for a
    given origin.
    
    Related to T2345



See https://jenkins.softwareheritage.org/job/DSCH/job/tests-on-diff/428/ for more details.
Harbormaster completed remote builds in B22859: Diff 21894.Aug 3 2021, 12:53 PM
vlorentz accepted this revision.Aug 3 2021, 1:08 PM
This revision is now accepted and ready to land.Aug 3 2021, 1:08 PM
Closed by commit rDSCH1bcf84d5e66d: Add a successive_visits counter to origin visit stats (authored by douardda, committed by ardumont).  ·  Explain WhyAug 4 2021, 10:06 AM
This revision was automatically updated to reflect the committed changes.
ardumont added a commit: rDSCH1bcf84d5e66d: Add a successive_visits counter to origin visit stats.
Revision Contents
Changeset List
PathSizeCoverage (All)Coverage (Touched)
sql/
updates/
4 lines--
swh/
scheduler/
8 lines100%
Loading...
2 lines100%
Loading...
sql/
2 lines--
tests/
34 lines100%
Loading...
Diff 21894
View Options
sql/updates/29.sql
Loading...
View Options
swh/scheduler/journal_client.py
Loading...
View Options
swh/scheduler/model.py
Loading...
View Options
swh/scheduler/sql/30-schema.sql
Loading...
View Options
swh/scheduler/tests/test_journal_client.py
Loading...
About · Developer information · Legal