Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[syncd.sh,pmon.service] Prevent pmon from starting ahead of syncd #8

Merged

Conversation

stephenxs
Copy link
Owner

- What I did
Prevent pmon from starting ahead of syncd by adding syncd.service as "After" of pmon.service.
This PR is one of the options that solve the comments of [syncd.sh] stop pmon ahead of syncd in flows except warm reboot #7

- How I did it
During system starting, pmon isn't supposed to start ahead of syncd starting in order to avoid racing condition between syncd and pmon.
Currently it is done by killing pmon if is alive when syncd is starting. However such implementation is still risky. Consider the following flow:

  1. pmon is inactive when syncd.sh is checking. but syncd.sh is scheduled out somehow just ahead of "chipdown" called
  2. systemd is switched in and starts pmon service
  3. at this point, pmon and syncd are running simultaneously, critical section broken and racing condition formed

To prevent that issue, ony solution is to add syncd as "After" in pmon.service, which ensure that whenever pmon starts syncd has been started.
However, dong so requires to defer starting pmon.service after syncd.service has fully started otherwise a deadlock is formed as following:

  1. syncd.sh starts pmon ahead of itself fully started, while
  2. pmon not being able to start due to syncd, one of its "After", not fully started.
  3. as a result, syncd and pmon have to wait for each other forever

To solve that, move starting pmon.service to "wait()" so that pmon is started after syncd fully started, breaking the deadlock.

- How to verify it

- Description for the changelog

- A picture of a cute animal (not mandatory but encouraged)

During system starting, pmon isn't supposed to start ahead of syncd starting in order to avoid racing condition between syncd and pmon.
Currently it is done by killing pmon if is alive when syncd is starting. However such implementation is still risky. Consider the following flow:
1. pmon is inactive when syncd.sh is checking. but syncd.sh is scheduled out somehow just ahead of "chipdown" called
2. systemd is switched in and starts pmon service
3. at this point, pmon and syncd are running simultaneously, critical section broken and racing condition formed
To prevent that issue, ony solution is to add syncd as "After" in pmon.service, which ensure that whenever pmon starts syncd has been started.
However, dong so requires to defer starting pmon.service after syncd.service has fully started otherwise a deadlock is formed as following:
1. syncd.sh starts pmon ahead of itself fully started, while
2. pmon not being able to start due to syncd, one of its "After", not fully started.
3. as a result, syncd and pmon have to wait for each other forever
To solve that, move starting pmon.service to "wait()" so that pmon is started after syncd fully started, breaking the deadlock.
@nazariig
Copy link

@stephenxs Maybe it is a good idea to have the entire synchronization flow based on systemd dependencies and remove all the rest stuff from syncd.sh? The pmon stop for warm-fast flows can be done in a dedicated scripts. Could you please check sych an option and compare it to existing approach? If you have any concerns we can discuss them.

@nazariig
Copy link

nazariig commented Sep 17, 2019

The idea to have Wants=pmon.service in syncd.service and After=syncd.service with Requires=syncd.service in pmon.service. This will definitely simplify the entire flow. What do you think?

@stephenxs stephenxs merged commit d5376e3 into pmon-dependency-syncd Sep 19, 2019
@stephenxs stephenxs deleted the pmon-dependency-syncd-defer-pmon-start branch September 19, 2019 14:24
stephenxs pushed a commit that referenced this pull request Apr 20, 2020
stephenxs pushed a commit that referenced this pull request Dec 10, 2020
This update brings in the following commits.

86c1108 Enable arm architecture to build in addition to amd64 (#37)
4acb2c3 fix bugs and enhance Transformer (#35)
49e5a22 ygot related enhancements and fixes (#34)
51224de Fix ietf yang search path for cvl schema builds (#32)
3c6cdb3 CVL Changes #8: 'must' and 'when' expression evaluation (#31)
dabf231 CVL Changes #7: 'leafref' evaluation (#28)
6f9535f CVL Changes #6: Customized Xpath Engine integration (#27)
5e2466b DB-Layer fixes/enhancements (#26)
9a27302 CVL Changes #4: Implementation of new CVL APIs (#22)
dbf1093 Translib support for authorization, yang versioning and Delete flag (#21)
80f369e CVL Changes #5: YParser enhancement (#23)
904ce18 CVL Changes #3: Multi-db instance support (#20)
9d24a34 CVL Changes #2:  YValidator infra changes for evaluating xpath expression (#19)
f3fc40f CVL Changes #1: Initial CVL code reorganization and common infra changes (#18)
4922601 Bulk and RPC API support in translib (#16)
1d730df RFC7895 yang module library implementation (#15)
stephenxs pushed a commit that referenced this pull request Nov 16, 2021
Update Barefoot platform support for Bullseye and 5.10 kernel, and add
python3-venv.
stephenxs pushed a commit that referenced this pull request Dec 14, 2021
#### What I did 

[sonic-linkmgrd][master] submodule update

6c6151b Fix unstable unit tests (state change handler wasn't invoked) (#8)
2f7dc0a support code diff coverage (#5)
83f0002 Force mux state switch to standby if triggered from Cli (#6)

signed-off-by: Jing Zhang [email protected]
stephenxs pushed a commit that referenced this pull request Dec 14, 2021
6c6151b Fix unstable unit tests (state change handler wasn't invoked) (#8)
2f7dc0a support code diff coverage (#5)
83f0002 Force mux state switch to standby if triggered from Cli (#6)

signed-off-by: Jing Zhang [email protected]
stephenxs pushed a commit that referenced this pull request Feb 10, 2022
* [BFN] Updated platform APIs impl

Signed-off-by: Andriy Kokhan <[email protected]>

* Extended BFN platform SFP APIs implementation

* Update sfp.py

* [BFN] Extended SFP platform plugin implementation

Signed-off-by: Andriy Kokhan <[email protected]>

* [BFN] Extended Fans platform plugin implementation

* [BFN] divided classes Fan and  FanDrawer into 2 files

* Signed-off-by: Vadym Yashchenko <[email protected]>

What I did
	Add get_model() function
	Add get_low_critical_threshold() function
	Change __get(...) function.
How I did it
	Differnece from previous implementation of __get(...) function is return real value or -9999.9 if value is not provided by thrift API

* Add get_presence() function and revised __get() function

Signed-off-by: Vadym Yashchenko <[email protected]>

* [BFN] Updated PSU platform APIs impl

Signed-off-by: Dmytro Lytvynenko <[email protected]>

* Added BFN PSU cache (#9)

Signed-off-by: Andriy Kokhan <[email protected]>

* [BFN]  Fans and Fantray platform APIs update (#7)

* [BFN] Updated SFP platform APIs (#10)

Signed-off-by: Volodymyr Boyko <[email protected]>

* [BFN] Updated platform API for thermal (#8)

* Signed-off-by: Vadym Yashchenko <[email protected]>

* Revert "[BFN]  Fans and Fantray platform APIs update (#7)" (#11)

This reverts commit c62a733.

* Add support health monitor system (#15)

Signed-off-by: Petro Bratash <[email protected]>

* Update chassis.py

* [BFN] Updated FANs and FAN Tray platform API (#14)

* Fix fix_alignment (#17)

Signed-off-by: Petro Bratash <[email protected]>

* [BFN] Improvement show environment (#16)

* Added PSU temperature skip into platform.json (#18)

Signed-off-by: Andriy Kokhan <[email protected]>

* Do not skip psud on Newport

Signed-off-by: Andriy Kokhan <[email protected]>

* [BFN] fix fan status from Not OK to Ok (#19)

* [BFN] Updated SFP platform plugin (#13)

Signed-off-by: Volodymyr Boyko <[email protected]>

* [DPB] Fix typo for Ethernet0 2x200G[100G,40G] breakout mode (#21)

Signed-off-by: Mykola Gerasymenko <[email protected]>

* [barefoot] Tmp fix vendor_rev (#22)

Signed-off-by: Volodymyr Boyko <[email protected]>

* Fixed python issues in sonic_platform/fan_drawer.py

Signed-off-by: Andriy Kokhan <[email protected]>

* Updated fan_drawer.py

* Fixing trailing white spaces in fan_drawer.py

* [BFN] Fix thrift for SFPs API

Signed-off-by: Volodymyr Boyko <[email protected]>

* In platform.json, replaced 'false' with '0' to workaround ast.literal_eval() issue

Signed-off-by: Andriy Kokhan <[email protected]>

* [Newport] Thermal manager  (#23)

* Signed-off-by: Vadym Yashchenko <[email protected]>

* Revert "In platform.json, replaced 'false' with '0' to workaround ast.literal_eval() issue"

This reverts commit 1e73127.

* Removed 'controllable' options from platform.json to fix factory default config generation

Signed-off-by: Andriy Kokhan <[email protected]>

* Update thermal_manager.py

* Migrated SFP plugin to sonic_xcvr API (#30)

Signed-off-by: Andriy Kokhan <[email protected]>

Co-authored-by: KostiantynYarovyiBf <[email protected]>
Co-authored-by: Vadym Yashchenko <[email protected]>
Co-authored-by: Dmytro Lytvynenko <[email protected]>
Co-authored-by: Volodymyr Boiko <[email protected]>
Co-authored-by: Petro Bratash <[email protected]>
Co-authored-by: Mykola Gerasymenko <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants